-
Notifications
You must be signed in to change notification settings - Fork 12.8k
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
Automatically enable cross-crate inlining for small functions #116505
Conversation
(rustbot has picked a reviewer for you, use r? to override) |
@bors try @rust-timer queue |
This comment has been minimized.
This comment has been minimized.
Automatically enable cross-crate inlining for small functions This is a work-in-progress. For example I have not thought at all about the cost model and I am sure that the threshold is too high. But I'm curious to know how this looks in perf. It certainly has some unique effects on codegen.
This comment has been minimized.
This comment has been minimized.
☀️ Try build successful - checks-actions |
This comment has been minimized.
This comment has been minimized.
Finished benchmarking commit (f1cbf12): comparison URL. Overall result: ❌✅ regressions and improvements - ACTION NEEDEDBenchmarking this pull request likely means that it is perf-sensitive, so we're automatically marking it as not fit for rolling up. While you can manually mark this PR as fit for rollup, we strongly recommend not doing so since this PR may lead to changes in compiler perf. Next Steps: If you can justify the regressions found in this try perf run, please indicate this with @bors rollup=never Instruction countThis is a highly reliable metric that was used to determine the overall result at the top of this comment.
Max RSS (memory usage)ResultsThis is a less reliable metric that may be of interest but was not used to determine the overall result at the top of this comment.
CyclesResultsThis is a less reliable metric that may be of interest but was not used to determine the overall result at the top of this comment.
Binary sizeResultsThis is a less reliable metric that may be of interest but was not used to determine the overall result at the top of this comment.
Bootstrap: 625.502s -> 742.509s (18.71%) |
This should be less silly than the last one because now we don't do the per-CGU thing for incr builds. And I also halved the default threshold. @bors try @rust-timer queue |
This comment has been minimized.
This comment has been minimized.
Automatically enable cross-crate inlining for small functions This is a work-in-progress. For example I have not thought at all about the cost model and I am sure that the threshold is too high. But I'm curious to know how this looks in perf. It certainly has some unique effects on codegen.
This comment has been minimized.
This comment has been minimized.
☀️ Try build successful - checks-actions |
This comment has been minimized.
This comment has been minimized.
Finished benchmarking commit (26bbeca): comparison URL. Overall result: ❌✅ regressions and improvements - ACTION NEEDEDBenchmarking this pull request likely means that it is perf-sensitive, so we're automatically marking it as not fit for rolling up. While you can manually mark this PR as fit for rollup, we strongly recommend not doing so since this PR may lead to changes in compiler perf. Next Steps: If you can justify the regressions found in this try perf run, please indicate this with @bors rollup=never Instruction countThis is a highly reliable metric that was used to determine the overall result at the top of this comment.
Max RSS (memory usage)ResultsThis is a less reliable metric that may be of interest but was not used to determine the overall result at the top of this comment.
CyclesResultsThis is a less reliable metric that may be of interest but was not used to determine the overall result at the top of this comment.
Binary sizeResultsThis is a less reliable metric that may be of interest but was not used to determine the overall result at the top of this comment.
Bootstrap: 622.673s -> 704.606s (13.16%) |
The culprit lied in rust-lang/rust#116505 In short, with default features turned off, main was trivial enough to be marked as inline function automatically which then made the symbol weak. Since nobody was referencing it, it got stripped away. Marking main in default-features=false config as #[inline(never)] or replacing 1 + 1 in it's body with a simple println!("foo") call (to make the main function sophisticated enough not to be subject to the new automatic marking as inline) makes the test pass again.
which, compared to the previous 1.70.0, has the following impacts on `zellij`: 1. [cross-crate inlining][8] This should increase application performance, as functions can now be inlined across crates. 2. [`async fn` in traits][9] This would allow us to drop the `async_trait` dependency, but it is currently still required by 3 other dependencies. Build time in debug mode (on my own PC) is cut down from 256s to 189s (for a clean build). Build time in release mode is cut down from 473s to 391s (for a clean build). Binary sizes only change minimally (825 MB -> 807 MB in debug, 29 MB -> 30 MB in release). [8]: rust-lang/rust#116505 [9]: rust-lang/rust#115822
* rust-toolchain: Bump toolchain version to 1.69.0 which, compared to the previous 1.67.0, has the following impacts on `zellij`: - [Turn off debuginfo for build deps][2]: Increases build time (on my machine) from ~230 s in 1.67.0 to ~250 s now, *which is unexpected* This version also changes [handling of the `default-features` flag][3] when specifying dependencies in `Cargo.toml`. If a dependent crate requires `default-features = true` on a crate that is required as `default-features = false` further up the dependency tree, the `true` setting "wins". We only specify `default-features = false` for three crates total: - `names`: This is used only by us - `surf`: This is used only by us - `vte`: This is also required by `strip-ansi-escapes`, but that has `default-features = false` as well How this affects our transitive dependencies is unknown at this point. [2]: rust-lang/cargo#11252 [3]: rust-lang/cargo#11409 * rust-toolchain: Bump toolchain version to 1.70.0 which, compared to the previous 1.69.0, as the following impacts on `zellij`: 1. [Enable sparse registry checkout for crates.io by default][1] This drastically increases the time to first build on a fresh rust installation/a rust installation with a clean cargo registry cache. Previously it took about 75s to populate the deps/cache (with `cargo fetch --locked` and ~100 MBit/s network), whereas now the same process takes ~10 s. 2. [The `OnceCell` type is now part of std][2] In theory, this would allow us to cut a dependency from `zellij-utils`, but the `once_cell` crate is pulled in by another 16 deps, so there's no point in attempting it right now. Build times and binary sizes are unaffected by this change compared to the previous 1.69.0 toolchain. [1]: rust-lang/cargo#11791 [2]: https://doc.rust-lang.org/stable/std/cell/struct.OnceCell.html * rust-toolchain: Bump toolchain version to 1.75.0 which, compared to the previous 1.70.0, has the following impacts on `zellij`: 1. [cross-crate inlining][8] This should increase application performance, as functions can now be inlined across crates. 2. [`async fn` in traits][9] This would allow us to drop the `async_trait` dependency, but it is currently still required by 3 other dependencies. Build time in debug mode (on my own PC) is cut down from 256s to 189s (for a clean build). Build time in release mode is cut down from 473s to 391s (for a clean build). Binary sizes only change minimally (825 MB -> 807 MB in debug, 29 MB -> 30 MB in release). [8]: rust-lang/rust#116505 [9]: rust-lang/rust#115822 * chore: Apply rustfmt. * CHANGELOG: Add PR #3039.
which, compared to the previous 1.70.0, has the following impacts on `zellij`: 1. [cross-crate inlining][8] This should increase application performance, as functions can now be inlined across crates. 2. [`async fn` in traits][9] This would allow us to drop the `async_trait` dependency, but it is currently still required by 3 other dependencies. Build time in debug mode (on my own PC) is cut down from 256s to 189s (for a clean build). Build time in release mode is cut down from 473s to 391s (for a clean build). Binary sizes only change minimally (825 MB -> 807 MB in debug, 29 MB -> 30 MB in release). [8]: rust-lang/rust#116505 [9]: rust-lang/rust#115822
With this change, is there any convenient way for a crate to indicate that it should be compiled with full optimizations but none of its public API should be inlined unless marked |
That attribute is documented as a hint, so no. I do not think we have a language feature that guarantees what you are asking for. It seems like a plausible thing to add but that's a bit outside my area. |
* rust-toolchain: Bump toolchain version to 1.69.0 which, compared to the previous 1.67.0, has the following impacts on `zellij`: - [Turn off debuginfo for build deps][2]: Increases build time (on my machine) from ~230 s in 1.67.0 to ~250 s now, *which is unexpected* This version also changes [handling of the `default-features` flag][3] when specifying dependencies in `Cargo.toml`. If a dependent crate requires `default-features = true` on a crate that is required as `default-features = false` further up the dependency tree, the `true` setting "wins". We only specify `default-features = false` for three crates total: - `names`: This is used only by us - `surf`: This is used only by us - `vte`: This is also required by `strip-ansi-escapes`, but that has `default-features = false` as well How this affects our transitive dependencies is unknown at this point. [2]: rust-lang/cargo#11252 [3]: rust-lang/cargo#11409 * rust-toolchain: Bump toolchain version to 1.70.0 which, compared to the previous 1.69.0, as the following impacts on `zellij`: 1. [Enable sparse registry checkout for crates.io by default][1] This drastically increases the time to first build on a fresh rust installation/a rust installation with a clean cargo registry cache. Previously it took about 75s to populate the deps/cache (with `cargo fetch --locked` and ~100 MBit/s network), whereas now the same process takes ~10 s. 2. [The `OnceCell` type is now part of std][2] In theory, this would allow us to cut a dependency from `zellij-utils`, but the `once_cell` crate is pulled in by another 16 deps, so there's no point in attempting it right now. Build times and binary sizes are unaffected by this change compared to the previous 1.69.0 toolchain. [1]: rust-lang/cargo#11791 [2]: https://doc.rust-lang.org/stable/std/cell/struct.OnceCell.html * rust-toolchain: Bump toolchain version to 1.75.0 which, compared to the previous 1.70.0, has the following impacts on `zellij`: 1. [cross-crate inlining][8] This should increase application performance, as functions can now be inlined across crates. 2. [`async fn` in traits][9] This would allow us to drop the `async_trait` dependency, but it is currently still required by 3 other dependencies. Build time in debug mode (on my own PC) is cut down from 256s to 189s (for a clean build). Build time in release mode is cut down from 473s to 391s (for a clean build). Binary sizes only change minimally (825 MB -> 807 MB in debug, 29 MB -> 30 MB in release). [8]: rust-lang/rust#116505 [9]: rust-lang/rust#115822 * chore: Apply rustfmt. * CHANGELOG: Add PR zellij-org#3039.
which, compared to the previous 1.70.0, has the following impacts on `zellij`: 1. [cross-crate inlining][8] This should increase application performance, as functions can now be inlined across crates. 2. [`async fn` in traits][9] This would allow us to drop the `async_trait` dependency, but it is currently still required by 3 other dependencies. Build time in debug mode (on my own PC) is cut down from 256s to 189s (for a clean build). Build time in release mode is cut down from 473s to 391s (for a clean build). Binary sizes only change minimally (825 MB -> 807 MB in debug, 29 MB -> 30 MB in release). [8]: rust-lang/rust#116505 [9]: rust-lang/rust#115822
* refactor: Simplify transfer_rows_from_viewport_to_lines_above next_lines is always consolidated to a single Row, which immediately gets removed - we can remove some dead code as a result * perf: Batch remove rows from the viewport for performance Given a 1MB line catted into the terminal, a toggle-fullscreen + toggle-fullscreen + close-pane + `run true` goes from ~9s to ~3s * perf: Optimize Row::drain_until by splitting chars in one step Given a 10MB line catted into the terminal, a toggle-fullscreen + toggle-fullscreen + close-pane + `run true` goes from ~23s to ~20s * refactor: Simplify `if let` into a `.map` * refactor: There are only new saved coordinates when there were old ones * refactor: Unify viewport transfer: use common variable names * fix: Use same saved cursor logic in height resize as width See #2182 for original introduction that only added it in one branch, this fixes an issue where the saved cursor was incorrectly reset when the real cursor was * fix: Correct saved+real cursor calculations when reflowing long lines * fix: Don't create canonical lines if cursor ends on EOL after resize Previously if a 20 character line were split into two 10 character lines, the cursor would be placed on the line after the two lines. New characters would then be treated as a new canonical line. This commit fixes this by biasing cursors to the end of the previous line. * fix: for cursor index calculation in lines that are already wrapped * chore: test for real/saved cursor position being handled separately * chore: Apply cargo format * chore(repo): update issue templates * Bump rust version to 1.75.0 (#3039) * rust-toolchain: Bump toolchain version to 1.69.0 which, compared to the previous 1.67.0, has the following impacts on `zellij`: - [Turn off debuginfo for build deps][2]: Increases build time (on my machine) from ~230 s in 1.67.0 to ~250 s now, *which is unexpected* This version also changes [handling of the `default-features` flag][3] when specifying dependencies in `Cargo.toml`. If a dependent crate requires `default-features = true` on a crate that is required as `default-features = false` further up the dependency tree, the `true` setting "wins". We only specify `default-features = false` for three crates total: - `names`: This is used only by us - `surf`: This is used only by us - `vte`: This is also required by `strip-ansi-escapes`, but that has `default-features = false` as well How this affects our transitive dependencies is unknown at this point. [2]: rust-lang/cargo#11252 [3]: rust-lang/cargo#11409 * rust-toolchain: Bump toolchain version to 1.70.0 which, compared to the previous 1.69.0, as the following impacts on `zellij`: 1. [Enable sparse registry checkout for crates.io by default][1] This drastically increases the time to first build on a fresh rust installation/a rust installation with a clean cargo registry cache. Previously it took about 75s to populate the deps/cache (with `cargo fetch --locked` and ~100 MBit/s network), whereas now the same process takes ~10 s. 2. [The `OnceCell` type is now part of std][2] In theory, this would allow us to cut a dependency from `zellij-utils`, but the `once_cell` crate is pulled in by another 16 deps, so there's no point in attempting it right now. Build times and binary sizes are unaffected by this change compared to the previous 1.69.0 toolchain. [1]: rust-lang/cargo#11791 [2]: https://doc.rust-lang.org/stable/std/cell/struct.OnceCell.html * rust-toolchain: Bump toolchain version to 1.75.0 which, compared to the previous 1.70.0, has the following impacts on `zellij`: 1. [cross-crate inlining][8] This should increase application performance, as functions can now be inlined across crates. 2. [`async fn` in traits][9] This would allow us to drop the `async_trait` dependency, but it is currently still required by 3 other dependencies. Build time in debug mode (on my own PC) is cut down from 256s to 189s (for a clean build). Build time in release mode is cut down from 473s to 391s (for a clean build). Binary sizes only change minimally (825 MB -> 807 MB in debug, 29 MB -> 30 MB in release). [8]: rust-lang/rust#116505 [9]: rust-lang/rust#115822 * chore: Apply rustfmt. * CHANGELOG: Add PR #3039. * feat(plugins): introduce 'pipes', allowing users to pipe data to and control plugins from the command line (#3066) * prototype - working with message from the cli * prototype - pipe from the CLI to plugins * prototype - pipe from the CLI to plugins and back again * prototype - working with better cli interface * prototype - working after removing unused stuff * prototype - working with launching plugin if it is not launched, also fixed event ordering * refactor: change message to cli-message * prototype - allow plugins to send messages to each other * fix: allow cli messages to send plugin parameters (and implement backpressure) * fix: use input_pipe_id to identify cli pipes instead of their message name * fix: come cleanups and add skip_cache parameter * fix: pipe/client-server communication robustness * fix: leaking messages between plugins while loading * feat: allow plugins to specify how a new plugin instance is launched when sending messages * fix: add permissions * refactor: adjust cli api * fix: improve cli plugin loading error messages * docs: cli pipe * fix: take plugin configuration into account when messaging between plugins * refactor: pipe message protobuf interface * refactor: update(event) -> pipe * refactor - rename CliMessage to CliPipe * fix: add is_private to pipes and change some naming * refactor - cli client * refactor: various cleanups * style(fmt): rustfmt * fix(pipes): backpressure across multiple plugins * style: some cleanups * style(fmt): rustfmt * style: fix merge conflict mistake * style(wording): clarify pipe permission * docs(changelog): introduce pipes * fix: add some robustness and future proofing * fix e2e tests --------- Co-authored-by: Aram Drevekenin <[email protected]> Co-authored-by: har7an <[email protected]>
* rust-toolchain: Bump toolchain version to 1.69.0 which, compared to the previous 1.67.0, has the following impacts on `zellij`: - [Turn off debuginfo for build deps][2]: Increases build time (on my machine) from ~230 s in 1.67.0 to ~250 s now, *which is unexpected* This version also changes [handling of the `default-features` flag][3] when specifying dependencies in `Cargo.toml`. If a dependent crate requires `default-features = true` on a crate that is required as `default-features = false` further up the dependency tree, the `true` setting "wins". We only specify `default-features = false` for three crates total: - `names`: This is used only by us - `surf`: This is used only by us - `vte`: This is also required by `strip-ansi-escapes`, but that has `default-features = false` as well How this affects our transitive dependencies is unknown at this point. [2]: rust-lang/cargo#11252 [3]: rust-lang/cargo#11409 * rust-toolchain: Bump toolchain version to 1.70.0 which, compared to the previous 1.69.0, as the following impacts on `zellij`: 1. [Enable sparse registry checkout for crates.io by default][1] This drastically increases the time to first build on a fresh rust installation/a rust installation with a clean cargo registry cache. Previously it took about 75s to populate the deps/cache (with `cargo fetch --locked` and ~100 MBit/s network), whereas now the same process takes ~10 s. 2. [The `OnceCell` type is now part of std][2] In theory, this would allow us to cut a dependency from `zellij-utils`, but the `once_cell` crate is pulled in by another 16 deps, so there's no point in attempting it right now. Build times and binary sizes are unaffected by this change compared to the previous 1.69.0 toolchain. [1]: rust-lang/cargo#11791 [2]: https://doc.rust-lang.org/stable/std/cell/struct.OnceCell.html * rust-toolchain: Bump toolchain version to 1.75.0 which, compared to the previous 1.70.0, has the following impacts on `zellij`: 1. [cross-crate inlining][8] This should increase application performance, as functions can now be inlined across crates. 2. [`async fn` in traits][9] This would allow us to drop the `async_trait` dependency, but it is currently still required by 3 other dependencies. Build time in debug mode (on my own PC) is cut down from 256s to 189s (for a clean build). Build time in release mode is cut down from 473s to 391s (for a clean build). Binary sizes only change minimally (825 MB -> 807 MB in debug, 29 MB -> 30 MB in release). [8]: rust-lang/rust#116505 [9]: rust-lang/rust#115822 * chore: Apply rustfmt. * CHANGELOG: Add PR zellij-org#3039.
…3032) * refactor: Simplify transfer_rows_from_viewport_to_lines_above next_lines is always consolidated to a single Row, which immediately gets removed - we can remove some dead code as a result * perf: Batch remove rows from the viewport for performance Given a 1MB line catted into the terminal, a toggle-fullscreen + toggle-fullscreen + close-pane + `run true` goes from ~9s to ~3s * perf: Optimize Row::drain_until by splitting chars in one step Given a 10MB line catted into the terminal, a toggle-fullscreen + toggle-fullscreen + close-pane + `run true` goes from ~23s to ~20s * refactor: Simplify `if let` into a `.map` * refactor: There are only new saved coordinates when there were old ones * refactor: Unify viewport transfer: use common variable names * fix: Use same saved cursor logic in height resize as width See zellij-org#2182 for original introduction that only added it in one branch, this fixes an issue where the saved cursor was incorrectly reset when the real cursor was * fix: Correct saved+real cursor calculations when reflowing long lines * fix: Don't create canonical lines if cursor ends on EOL after resize Previously if a 20 character line were split into two 10 character lines, the cursor would be placed on the line after the two lines. New characters would then be treated as a new canonical line. This commit fixes this by biasing cursors to the end of the previous line. * fix: for cursor index calculation in lines that are already wrapped * chore: test for real/saved cursor position being handled separately * chore: Apply cargo format * chore(repo): update issue templates * Bump rust version to 1.75.0 (zellij-org#3039) * rust-toolchain: Bump toolchain version to 1.69.0 which, compared to the previous 1.67.0, has the following impacts on `zellij`: - [Turn off debuginfo for build deps][2]: Increases build time (on my machine) from ~230 s in 1.67.0 to ~250 s now, *which is unexpected* This version also changes [handling of the `default-features` flag][3] when specifying dependencies in `Cargo.toml`. If a dependent crate requires `default-features = true` on a crate that is required as `default-features = false` further up the dependency tree, the `true` setting "wins". We only specify `default-features = false` for three crates total: - `names`: This is used only by us - `surf`: This is used only by us - `vte`: This is also required by `strip-ansi-escapes`, but that has `default-features = false` as well How this affects our transitive dependencies is unknown at this point. [2]: rust-lang/cargo#11252 [3]: rust-lang/cargo#11409 * rust-toolchain: Bump toolchain version to 1.70.0 which, compared to the previous 1.69.0, as the following impacts on `zellij`: 1. [Enable sparse registry checkout for crates.io by default][1] This drastically increases the time to first build on a fresh rust installation/a rust installation with a clean cargo registry cache. Previously it took about 75s to populate the deps/cache (with `cargo fetch --locked` and ~100 MBit/s network), whereas now the same process takes ~10 s. 2. [The `OnceCell` type is now part of std][2] In theory, this would allow us to cut a dependency from `zellij-utils`, but the `once_cell` crate is pulled in by another 16 deps, so there's no point in attempting it right now. Build times and binary sizes are unaffected by this change compared to the previous 1.69.0 toolchain. [1]: rust-lang/cargo#11791 [2]: https://doc.rust-lang.org/stable/std/cell/struct.OnceCell.html * rust-toolchain: Bump toolchain version to 1.75.0 which, compared to the previous 1.70.0, has the following impacts on `zellij`: 1. [cross-crate inlining][8] This should increase application performance, as functions can now be inlined across crates. 2. [`async fn` in traits][9] This would allow us to drop the `async_trait` dependency, but it is currently still required by 3 other dependencies. Build time in debug mode (on my own PC) is cut down from 256s to 189s (for a clean build). Build time in release mode is cut down from 473s to 391s (for a clean build). Binary sizes only change minimally (825 MB -> 807 MB in debug, 29 MB -> 30 MB in release). [8]: rust-lang/rust#116505 [9]: rust-lang/rust#115822 * chore: Apply rustfmt. * CHANGELOG: Add PR zellij-org#3039. * feat(plugins): introduce 'pipes', allowing users to pipe data to and control plugins from the command line (zellij-org#3066) * prototype - working with message from the cli * prototype - pipe from the CLI to plugins * prototype - pipe from the CLI to plugins and back again * prototype - working with better cli interface * prototype - working after removing unused stuff * prototype - working with launching plugin if it is not launched, also fixed event ordering * refactor: change message to cli-message * prototype - allow plugins to send messages to each other * fix: allow cli messages to send plugin parameters (and implement backpressure) * fix: use input_pipe_id to identify cli pipes instead of their message name * fix: come cleanups and add skip_cache parameter * fix: pipe/client-server communication robustness * fix: leaking messages between plugins while loading * feat: allow plugins to specify how a new plugin instance is launched when sending messages * fix: add permissions * refactor: adjust cli api * fix: improve cli plugin loading error messages * docs: cli pipe * fix: take plugin configuration into account when messaging between plugins * refactor: pipe message protobuf interface * refactor: update(event) -> pipe * refactor - rename CliMessage to CliPipe * fix: add is_private to pipes and change some naming * refactor - cli client * refactor: various cleanups * style(fmt): rustfmt * fix(pipes): backpressure across multiple plugins * style: some cleanups * style(fmt): rustfmt * style: fix merge conflict mistake * style(wording): clarify pipe permission * docs(changelog): introduce pipes * fix: add some robustness and future proofing * fix e2e tests --------- Co-authored-by: Aram Drevekenin <[email protected]> Co-authored-by: har7an <[email protected]>
… reducing size of TerminalCharacter (#3043) * refactor: Simplify transfer_rows_from_viewport_to_lines_above next_lines is always consolidated to a single Row, which immediately gets removed - we can remove some dead code as a result * perf: Batch remove rows from the viewport for performance Given a 1MB line catted into the terminal, a toggle-fullscreen + toggle-fullscreen + close-pane + `run true` goes from ~9s to ~3s * perf: Optimize Row::drain_until by splitting chars in one step Given a 10MB line catted into the terminal, a toggle-fullscreen + toggle-fullscreen + close-pane + `run true` goes from ~23s to ~20s * refactor: Simplify `if let` into a `.map` * refactor: There are only new saved coordinates when there were old ones * refactor: Unify viewport transfer: use common variable names * fix: Use same saved cursor logic in height resize as width See #2182 for original introduction that only added it in one branch, this fixes an issue where the saved cursor was incorrectly reset when the real cursor was * fix: Correct saved+real cursor calculations when reflowing long lines * fix: Don't create canonical lines if cursor ends on EOL after resize Previously if a 20 character line were split into two 10 character lines, the cursor would be placed on the line after the two lines. New characters would then be treated as a new canonical line. This commit fixes this by biasing cursors to the end of the previous line. * fix: for cursor index calculation in lines that are already wrapped * chore: test for real/saved cursor position being handled separately * chore: Apply cargo format * refactor: Unify viewport transfer: transfer + cursor update together * perf: Reduce size of TerminalCharacter from 72 to 60 bytes With a 10MB single line catted into a fresh terminal, VmRSS goes from 964092 kB to 874020 kB (as reported by /proc/<pid>/status) before/after this patch Given a 10MB line catted into the terminal, a toggle-fullscreen + toggle-fullscreen + close-pane + `run true` goes from ~15s to ~12.5s * fix(build): Don't unconditionally rebuild zellij-utils * refactor: Remove Copy impl on TerminalCharacter * perf: Rc styles to reduce TerminalCharacter from 60 to 24 bytes With a 10MB single line catted into a fresh terminal, VmRSS goes from 845156 kB to 478396 kB (as reported by /proc/<pid>/status) before/after this patch Given a 10MB line catted into the terminal, a toggle-fullscreen + toggle-fullscreen + close-pane + `run true` goes from ~12.5s to ~7s * perf: Remove RcCharacterStyles::Default, allow enum niche optimisation This reduces TerminalCharacter from 24 to 16 bytes With a 10MB single line catted into a fresh terminal, VmRSS goes from 478396 kB to 398108 kB (as reported by /proc/<pid>/status) before/after this patch Given a 10MB line catted into the terminal, a toggle-fullscreen + toggle-fullscreen + close-pane + `run true` goes from ~7s to ~4.75s * docs: link anchor omission from reset_all is deliberate reset_all is only used from ansi params, and ansi params don't control link anchor * fix: Remove no-op on variable that gets immediately dropped * refactor: Simplify replace_character_at logic The original condition checked absolute_x_index was in bounds, then used the index to manipulate it. This is equivalent to getting a ref to the character at that position and manipulating directly * chore: Run xtask format * chore(repo): update issue templates * Bump rust version to 1.75.0 (#3039) * rust-toolchain: Bump toolchain version to 1.69.0 which, compared to the previous 1.67.0, has the following impacts on `zellij`: - [Turn off debuginfo for build deps][2]: Increases build time (on my machine) from ~230 s in 1.67.0 to ~250 s now, *which is unexpected* This version also changes [handling of the `default-features` flag][3] when specifying dependencies in `Cargo.toml`. If a dependent crate requires `default-features = true` on a crate that is required as `default-features = false` further up the dependency tree, the `true` setting "wins". We only specify `default-features = false` for three crates total: - `names`: This is used only by us - `surf`: This is used only by us - `vte`: This is also required by `strip-ansi-escapes`, but that has `default-features = false` as well How this affects our transitive dependencies is unknown at this point. [2]: rust-lang/cargo#11252 [3]: rust-lang/cargo#11409 * rust-toolchain: Bump toolchain version to 1.70.0 which, compared to the previous 1.69.0, as the following impacts on `zellij`: 1. [Enable sparse registry checkout for crates.io by default][1] This drastically increases the time to first build on a fresh rust installation/a rust installation with a clean cargo registry cache. Previously it took about 75s to populate the deps/cache (with `cargo fetch --locked` and ~100 MBit/s network), whereas now the same process takes ~10 s. 2. [The `OnceCell` type is now part of std][2] In theory, this would allow us to cut a dependency from `zellij-utils`, but the `once_cell` crate is pulled in by another 16 deps, so there's no point in attempting it right now. Build times and binary sizes are unaffected by this change compared to the previous 1.69.0 toolchain. [1]: rust-lang/cargo#11791 [2]: https://doc.rust-lang.org/stable/std/cell/struct.OnceCell.html * rust-toolchain: Bump toolchain version to 1.75.0 which, compared to the previous 1.70.0, has the following impacts on `zellij`: 1. [cross-crate inlining][8] This should increase application performance, as functions can now be inlined across crates. 2. [`async fn` in traits][9] This would allow us to drop the `async_trait` dependency, but it is currently still required by 3 other dependencies. Build time in debug mode (on my own PC) is cut down from 256s to 189s (for a clean build). Build time in release mode is cut down from 473s to 391s (for a clean build). Binary sizes only change minimally (825 MB -> 807 MB in debug, 29 MB -> 30 MB in release). [8]: rust-lang/rust#116505 [9]: rust-lang/rust#115822 * chore: Apply rustfmt. * CHANGELOG: Add PR #3039. * feat(plugins): introduce 'pipes', allowing users to pipe data to and control plugins from the command line (#3066) * prototype - working with message from the cli * prototype - pipe from the CLI to plugins * prototype - pipe from the CLI to plugins and back again * prototype - working with better cli interface * prototype - working after removing unused stuff * prototype - working with launching plugin if it is not launched, also fixed event ordering * refactor: change message to cli-message * prototype - allow plugins to send messages to each other * fix: allow cli messages to send plugin parameters (and implement backpressure) * fix: use input_pipe_id to identify cli pipes instead of their message name * fix: come cleanups and add skip_cache parameter * fix: pipe/client-server communication robustness * fix: leaking messages between plugins while loading * feat: allow plugins to specify how a new plugin instance is launched when sending messages * fix: add permissions * refactor: adjust cli api * fix: improve cli plugin loading error messages * docs: cli pipe * fix: take plugin configuration into account when messaging between plugins * refactor: pipe message protobuf interface * refactor: update(event) -> pipe * refactor - rename CliMessage to CliPipe * fix: add is_private to pipes and change some naming * refactor - cli client * refactor: various cleanups * style(fmt): rustfmt * fix(pipes): backpressure across multiple plugins * style: some cleanups * style(fmt): rustfmt * style: fix merge conflict mistake * style(wording): clarify pipe permission * docs(changelog): introduce pipes * xtask: Disable pusing during publish (#3040) * xtask: Add `--no-push` flag to `publish` which can be used when simulating releases to work without a writable git fork of the zellij code. * xtask: Fix borrow issues * xtask/pipe: Require lockfile in publish to avoid errors from invalid dependency versions. * CHANGELOG: Add PR #3040. * fix(terminal): some real/saved cursor bugs during resize (#3032) * refactor: Simplify transfer_rows_from_viewport_to_lines_above next_lines is always consolidated to a single Row, which immediately gets removed - we can remove some dead code as a result * perf: Batch remove rows from the viewport for performance Given a 1MB line catted into the terminal, a toggle-fullscreen + toggle-fullscreen + close-pane + `run true` goes from ~9s to ~3s * perf: Optimize Row::drain_until by splitting chars in one step Given a 10MB line catted into the terminal, a toggle-fullscreen + toggle-fullscreen + close-pane + `run true` goes from ~23s to ~20s * refactor: Simplify `if let` into a `.map` * refactor: There are only new saved coordinates when there were old ones * refactor: Unify viewport transfer: use common variable names * fix: Use same saved cursor logic in height resize as width See #2182 for original introduction that only added it in one branch, this fixes an issue where the saved cursor was incorrectly reset when the real cursor was * fix: Correct saved+real cursor calculations when reflowing long lines * fix: Don't create canonical lines if cursor ends on EOL after resize Previously if a 20 character line were split into two 10 character lines, the cursor would be placed on the line after the two lines. New characters would then be treated as a new canonical line. This commit fixes this by biasing cursors to the end of the previous line. * fix: for cursor index calculation in lines that are already wrapped * chore: test for real/saved cursor position being handled separately * chore: Apply cargo format * chore(repo): update issue templates * Bump rust version to 1.75.0 (#3039) * rust-toolchain: Bump toolchain version to 1.69.0 which, compared to the previous 1.67.0, has the following impacts on `zellij`: - [Turn off debuginfo for build deps][2]: Increases build time (on my machine) from ~230 s in 1.67.0 to ~250 s now, *which is unexpected* This version also changes [handling of the `default-features` flag][3] when specifying dependencies in `Cargo.toml`. If a dependent crate requires `default-features = true` on a crate that is required as `default-features = false` further up the dependency tree, the `true` setting "wins". We only specify `default-features = false` for three crates total: - `names`: This is used only by us - `surf`: This is used only by us - `vte`: This is also required by `strip-ansi-escapes`, but that has `default-features = false` as well How this affects our transitive dependencies is unknown at this point. [2]: rust-lang/cargo#11252 [3]: rust-lang/cargo#11409 * rust-toolchain: Bump toolchain version to 1.70.0 which, compared to the previous 1.69.0, as the following impacts on `zellij`: 1. [Enable sparse registry checkout for crates.io by default][1] This drastically increases the time to first build on a fresh rust installation/a rust installation with a clean cargo registry cache. Previously it took about 75s to populate the deps/cache (with `cargo fetch --locked` and ~100 MBit/s network), whereas now the same process takes ~10 s. 2. [The `OnceCell` type is now part of std][2] In theory, this would allow us to cut a dependency from `zellij-utils`, but the `once_cell` crate is pulled in by another 16 deps, so there's no point in attempting it right now. Build times and binary sizes are unaffected by this change compared to the previous 1.69.0 toolchain. [1]: rust-lang/cargo#11791 [2]: https://doc.rust-lang.org/stable/std/cell/struct.OnceCell.html * rust-toolchain: Bump toolchain version to 1.75.0 which, compared to the previous 1.70.0, has the following impacts on `zellij`: 1. [cross-crate inlining][8] This should increase application performance, as functions can now be inlined across crates. 2. [`async fn` in traits][9] This would allow us to drop the `async_trait` dependency, but it is currently still required by 3 other dependencies. Build time in debug mode (on my own PC) is cut down from 256s to 189s (for a clean build). Build time in release mode is cut down from 473s to 391s (for a clean build). Binary sizes only change minimally (825 MB -> 807 MB in debug, 29 MB -> 30 MB in release). [8]: rust-lang/rust#116505 [9]: rust-lang/rust#115822 * chore: Apply rustfmt. * CHANGELOG: Add PR #3039. * feat(plugins): introduce 'pipes', allowing users to pipe data to and control plugins from the command line (#3066) * prototype - working with message from the cli * prototype - pipe from the CLI to plugins * prototype - pipe from the CLI to plugins and back again * prototype - working with better cli interface * prototype - working after removing unused stuff * prototype - working with launching plugin if it is not launched, also fixed event ordering * refactor: change message to cli-message * prototype - allow plugins to send messages to each other * fix: allow cli messages to send plugin parameters (and implement backpressure) * fix: use input_pipe_id to identify cli pipes instead of their message name * fix: come cleanups and add skip_cache parameter * fix: pipe/client-server communication robustness * fix: leaking messages between plugins while loading * feat: allow plugins to specify how a new plugin instance is launched when sending messages * fix: add permissions * refactor: adjust cli api * fix: improve cli plugin loading error messages * docs: cli pipe * fix: take plugin configuration into account when messaging between plugins * refactor: pipe message protobuf interface * refactor: update(event) -> pipe * refactor - rename CliMessage to CliPipe * fix: add is_private to pipes and change some naming * refactor - cli client * refactor: various cleanups * style(fmt): rustfmt * fix(pipes): backpressure across multiple plugins * style: some cleanups * style(fmt): rustfmt * style: fix merge conflict mistake * style(wording): clarify pipe permission * docs(changelog): introduce pipes * fix: add some robustness and future proofing * fix e2e tests --------- Co-authored-by: Aram Drevekenin <[email protected]> Co-authored-by: har7an <[email protected]> * fix integer overflow again (oops) --------- Co-authored-by: Aram Drevekenin <[email protected]> Co-authored-by: har7an <[email protected]>
* rust-toolchain: Bump toolchain version to 1.69.0 which, compared to the previous 1.67.0, has the following impacts on `zellij`: - [Turn off debuginfo for build deps][2]: Increases build time (on my machine) from ~230 s in 1.67.0 to ~250 s now, *which is unexpected* This version also changes [handling of the `default-features` flag][3] when specifying dependencies in `Cargo.toml`. If a dependent crate requires `default-features = true` on a crate that is required as `default-features = false` further up the dependency tree, the `true` setting "wins". We only specify `default-features = false` for three crates total: - `names`: This is used only by us - `surf`: This is used only by us - `vte`: This is also required by `strip-ansi-escapes`, but that has `default-features = false` as well How this affects our transitive dependencies is unknown at this point. [2]: rust-lang/cargo#11252 [3]: rust-lang/cargo#11409 * rust-toolchain: Bump toolchain version to 1.70.0 which, compared to the previous 1.69.0, as the following impacts on `zellij`: 1. [Enable sparse registry checkout for crates.io by default][1] This drastically increases the time to first build on a fresh rust installation/a rust installation with a clean cargo registry cache. Previously it took about 75s to populate the deps/cache (with `cargo fetch --locked` and ~100 MBit/s network), whereas now the same process takes ~10 s. 2. [The `OnceCell` type is now part of std][2] In theory, this would allow us to cut a dependency from `zellij-utils`, but the `once_cell` crate is pulled in by another 16 deps, so there's no point in attempting it right now. Build times and binary sizes are unaffected by this change compared to the previous 1.69.0 toolchain. [1]: rust-lang/cargo#11791 [2]: https://doc.rust-lang.org/stable/std/cell/struct.OnceCell.html * rust-toolchain: Bump toolchain version to 1.75.0 which, compared to the previous 1.70.0, has the following impacts on `zellij`: 1. [cross-crate inlining][8] This should increase application performance, as functions can now be inlined across crates. 2. [`async fn` in traits][9] This would allow us to drop the `async_trait` dependency, but it is currently still required by 3 other dependencies. Build time in debug mode (on my own PC) is cut down from 256s to 189s (for a clean build). Build time in release mode is cut down from 473s to 391s (for a clean build). Binary sizes only change minimally (825 MB -> 807 MB in debug, 29 MB -> 30 MB in release). [8]: rust-lang/rust#116505 [9]: rust-lang/rust#115822 * chore: Apply rustfmt. * CHANGELOG: Add PR zellij-org#3039.
…3032) * refactor: Simplify transfer_rows_from_viewport_to_lines_above next_lines is always consolidated to a single Row, which immediately gets removed - we can remove some dead code as a result * perf: Batch remove rows from the viewport for performance Given a 1MB line catted into the terminal, a toggle-fullscreen + toggle-fullscreen + close-pane + `run true` goes from ~9s to ~3s * perf: Optimize Row::drain_until by splitting chars in one step Given a 10MB line catted into the terminal, a toggle-fullscreen + toggle-fullscreen + close-pane + `run true` goes from ~23s to ~20s * refactor: Simplify `if let` into a `.map` * refactor: There are only new saved coordinates when there were old ones * refactor: Unify viewport transfer: use common variable names * fix: Use same saved cursor logic in height resize as width See zellij-org#2182 for original introduction that only added it in one branch, this fixes an issue where the saved cursor was incorrectly reset when the real cursor was * fix: Correct saved+real cursor calculations when reflowing long lines * fix: Don't create canonical lines if cursor ends on EOL after resize Previously if a 20 character line were split into two 10 character lines, the cursor would be placed on the line after the two lines. New characters would then be treated as a new canonical line. This commit fixes this by biasing cursors to the end of the previous line. * fix: for cursor index calculation in lines that are already wrapped * chore: test for real/saved cursor position being handled separately * chore: Apply cargo format * chore(repo): update issue templates * Bump rust version to 1.75.0 (zellij-org#3039) * rust-toolchain: Bump toolchain version to 1.69.0 which, compared to the previous 1.67.0, has the following impacts on `zellij`: - [Turn off debuginfo for build deps][2]: Increases build time (on my machine) from ~230 s in 1.67.0 to ~250 s now, *which is unexpected* This version also changes [handling of the `default-features` flag][3] when specifying dependencies in `Cargo.toml`. If a dependent crate requires `default-features = true` on a crate that is required as `default-features = false` further up the dependency tree, the `true` setting "wins". We only specify `default-features = false` for three crates total: - `names`: This is used only by us - `surf`: This is used only by us - `vte`: This is also required by `strip-ansi-escapes`, but that has `default-features = false` as well How this affects our transitive dependencies is unknown at this point. [2]: rust-lang/cargo#11252 [3]: rust-lang/cargo#11409 * rust-toolchain: Bump toolchain version to 1.70.0 which, compared to the previous 1.69.0, as the following impacts on `zellij`: 1. [Enable sparse registry checkout for crates.io by default][1] This drastically increases the time to first build on a fresh rust installation/a rust installation with a clean cargo registry cache. Previously it took about 75s to populate the deps/cache (with `cargo fetch --locked` and ~100 MBit/s network), whereas now the same process takes ~10 s. 2. [The `OnceCell` type is now part of std][2] In theory, this would allow us to cut a dependency from `zellij-utils`, but the `once_cell` crate is pulled in by another 16 deps, so there's no point in attempting it right now. Build times and binary sizes are unaffected by this change compared to the previous 1.69.0 toolchain. [1]: rust-lang/cargo#11791 [2]: https://doc.rust-lang.org/stable/std/cell/struct.OnceCell.html * rust-toolchain: Bump toolchain version to 1.75.0 which, compared to the previous 1.70.0, has the following impacts on `zellij`: 1. [cross-crate inlining][8] This should increase application performance, as functions can now be inlined across crates. 2. [`async fn` in traits][9] This would allow us to drop the `async_trait` dependency, but it is currently still required by 3 other dependencies. Build time in debug mode (on my own PC) is cut down from 256s to 189s (for a clean build). Build time in release mode is cut down from 473s to 391s (for a clean build). Binary sizes only change minimally (825 MB -> 807 MB in debug, 29 MB -> 30 MB in release). [8]: rust-lang/rust#116505 [9]: rust-lang/rust#115822 * chore: Apply rustfmt. * CHANGELOG: Add PR zellij-org#3039. * feat(plugins): introduce 'pipes', allowing users to pipe data to and control plugins from the command line (zellij-org#3066) * prototype - working with message from the cli * prototype - pipe from the CLI to plugins * prototype - pipe from the CLI to plugins and back again * prototype - working with better cli interface * prototype - working after removing unused stuff * prototype - working with launching plugin if it is not launched, also fixed event ordering * refactor: change message to cli-message * prototype - allow plugins to send messages to each other * fix: allow cli messages to send plugin parameters (and implement backpressure) * fix: use input_pipe_id to identify cli pipes instead of their message name * fix: come cleanups and add skip_cache parameter * fix: pipe/client-server communication robustness * fix: leaking messages between plugins while loading * feat: allow plugins to specify how a new plugin instance is launched when sending messages * fix: add permissions * refactor: adjust cli api * fix: improve cli plugin loading error messages * docs: cli pipe * fix: take plugin configuration into account when messaging between plugins * refactor: pipe message protobuf interface * refactor: update(event) -> pipe * refactor - rename CliMessage to CliPipe * fix: add is_private to pipes and change some naming * refactor - cli client * refactor: various cleanups * style(fmt): rustfmt * fix(pipes): backpressure across multiple plugins * style: some cleanups * style(fmt): rustfmt * style: fix merge conflict mistake * style(wording): clarify pipe permission * docs(changelog): introduce pipes * fix: add some robustness and future proofing * fix e2e tests --------- Co-authored-by: Aram Drevekenin <[email protected]> Co-authored-by: har7an <[email protected]>
… reducing size of TerminalCharacter (zellij-org#3043) * refactor: Simplify transfer_rows_from_viewport_to_lines_above next_lines is always consolidated to a single Row, which immediately gets removed - we can remove some dead code as a result * perf: Batch remove rows from the viewport for performance Given a 1MB line catted into the terminal, a toggle-fullscreen + toggle-fullscreen + close-pane + `run true` goes from ~9s to ~3s * perf: Optimize Row::drain_until by splitting chars in one step Given a 10MB line catted into the terminal, a toggle-fullscreen + toggle-fullscreen + close-pane + `run true` goes from ~23s to ~20s * refactor: Simplify `if let` into a `.map` * refactor: There are only new saved coordinates when there were old ones * refactor: Unify viewport transfer: use common variable names * fix: Use same saved cursor logic in height resize as width See zellij-org#2182 for original introduction that only added it in one branch, this fixes an issue where the saved cursor was incorrectly reset when the real cursor was * fix: Correct saved+real cursor calculations when reflowing long lines * fix: Don't create canonical lines if cursor ends on EOL after resize Previously if a 20 character line were split into two 10 character lines, the cursor would be placed on the line after the two lines. New characters would then be treated as a new canonical line. This commit fixes this by biasing cursors to the end of the previous line. * fix: for cursor index calculation in lines that are already wrapped * chore: test for real/saved cursor position being handled separately * chore: Apply cargo format * refactor: Unify viewport transfer: transfer + cursor update together * perf: Reduce size of TerminalCharacter from 72 to 60 bytes With a 10MB single line catted into a fresh terminal, VmRSS goes from 964092 kB to 874020 kB (as reported by /proc/<pid>/status) before/after this patch Given a 10MB line catted into the terminal, a toggle-fullscreen + toggle-fullscreen + close-pane + `run true` goes from ~15s to ~12.5s * fix(build): Don't unconditionally rebuild zellij-utils * refactor: Remove Copy impl on TerminalCharacter * perf: Rc styles to reduce TerminalCharacter from 60 to 24 bytes With a 10MB single line catted into a fresh terminal, VmRSS goes from 845156 kB to 478396 kB (as reported by /proc/<pid>/status) before/after this patch Given a 10MB line catted into the terminal, a toggle-fullscreen + toggle-fullscreen + close-pane + `run true` goes from ~12.5s to ~7s * perf: Remove RcCharacterStyles::Default, allow enum niche optimisation This reduces TerminalCharacter from 24 to 16 bytes With a 10MB single line catted into a fresh terminal, VmRSS goes from 478396 kB to 398108 kB (as reported by /proc/<pid>/status) before/after this patch Given a 10MB line catted into the terminal, a toggle-fullscreen + toggle-fullscreen + close-pane + `run true` goes from ~7s to ~4.75s * docs: link anchor omission from reset_all is deliberate reset_all is only used from ansi params, and ansi params don't control link anchor * fix: Remove no-op on variable that gets immediately dropped * refactor: Simplify replace_character_at logic The original condition checked absolute_x_index was in bounds, then used the index to manipulate it. This is equivalent to getting a ref to the character at that position and manipulating directly * chore: Run xtask format * chore(repo): update issue templates * Bump rust version to 1.75.0 (zellij-org#3039) * rust-toolchain: Bump toolchain version to 1.69.0 which, compared to the previous 1.67.0, has the following impacts on `zellij`: - [Turn off debuginfo for build deps][2]: Increases build time (on my machine) from ~230 s in 1.67.0 to ~250 s now, *which is unexpected* This version also changes [handling of the `default-features` flag][3] when specifying dependencies in `Cargo.toml`. If a dependent crate requires `default-features = true` on a crate that is required as `default-features = false` further up the dependency tree, the `true` setting "wins". We only specify `default-features = false` for three crates total: - `names`: This is used only by us - `surf`: This is used only by us - `vte`: This is also required by `strip-ansi-escapes`, but that has `default-features = false` as well How this affects our transitive dependencies is unknown at this point. [2]: rust-lang/cargo#11252 [3]: rust-lang/cargo#11409 * rust-toolchain: Bump toolchain version to 1.70.0 which, compared to the previous 1.69.0, as the following impacts on `zellij`: 1. [Enable sparse registry checkout for crates.io by default][1] This drastically increases the time to first build on a fresh rust installation/a rust installation with a clean cargo registry cache. Previously it took about 75s to populate the deps/cache (with `cargo fetch --locked` and ~100 MBit/s network), whereas now the same process takes ~10 s. 2. [The `OnceCell` type is now part of std][2] In theory, this would allow us to cut a dependency from `zellij-utils`, but the `once_cell` crate is pulled in by another 16 deps, so there's no point in attempting it right now. Build times and binary sizes are unaffected by this change compared to the previous 1.69.0 toolchain. [1]: rust-lang/cargo#11791 [2]: https://doc.rust-lang.org/stable/std/cell/struct.OnceCell.html * rust-toolchain: Bump toolchain version to 1.75.0 which, compared to the previous 1.70.0, has the following impacts on `zellij`: 1. [cross-crate inlining][8] This should increase application performance, as functions can now be inlined across crates. 2. [`async fn` in traits][9] This would allow us to drop the `async_trait` dependency, but it is currently still required by 3 other dependencies. Build time in debug mode (on my own PC) is cut down from 256s to 189s (for a clean build). Build time in release mode is cut down from 473s to 391s (for a clean build). Binary sizes only change minimally (825 MB -> 807 MB in debug, 29 MB -> 30 MB in release). [8]: rust-lang/rust#116505 [9]: rust-lang/rust#115822 * chore: Apply rustfmt. * CHANGELOG: Add PR zellij-org#3039. * feat(plugins): introduce 'pipes', allowing users to pipe data to and control plugins from the command line (zellij-org#3066) * prototype - working with message from the cli * prototype - pipe from the CLI to plugins * prototype - pipe from the CLI to plugins and back again * prototype - working with better cli interface * prototype - working after removing unused stuff * prototype - working with launching plugin if it is not launched, also fixed event ordering * refactor: change message to cli-message * prototype - allow plugins to send messages to each other * fix: allow cli messages to send plugin parameters (and implement backpressure) * fix: use input_pipe_id to identify cli pipes instead of their message name * fix: come cleanups and add skip_cache parameter * fix: pipe/client-server communication robustness * fix: leaking messages between plugins while loading * feat: allow plugins to specify how a new plugin instance is launched when sending messages * fix: add permissions * refactor: adjust cli api * fix: improve cli plugin loading error messages * docs: cli pipe * fix: take plugin configuration into account when messaging between plugins * refactor: pipe message protobuf interface * refactor: update(event) -> pipe * refactor - rename CliMessage to CliPipe * fix: add is_private to pipes and change some naming * refactor - cli client * refactor: various cleanups * style(fmt): rustfmt * fix(pipes): backpressure across multiple plugins * style: some cleanups * style(fmt): rustfmt * style: fix merge conflict mistake * style(wording): clarify pipe permission * docs(changelog): introduce pipes * xtask: Disable pusing during publish (zellij-org#3040) * xtask: Add `--no-push` flag to `publish` which can be used when simulating releases to work without a writable git fork of the zellij code. * xtask: Fix borrow issues * xtask/pipe: Require lockfile in publish to avoid errors from invalid dependency versions. * CHANGELOG: Add PR zellij-org#3040. * fix(terminal): some real/saved cursor bugs during resize (zellij-org#3032) * refactor: Simplify transfer_rows_from_viewport_to_lines_above next_lines is always consolidated to a single Row, which immediately gets removed - we can remove some dead code as a result * perf: Batch remove rows from the viewport for performance Given a 1MB line catted into the terminal, a toggle-fullscreen + toggle-fullscreen + close-pane + `run true` goes from ~9s to ~3s * perf: Optimize Row::drain_until by splitting chars in one step Given a 10MB line catted into the terminal, a toggle-fullscreen + toggle-fullscreen + close-pane + `run true` goes from ~23s to ~20s * refactor: Simplify `if let` into a `.map` * refactor: There are only new saved coordinates when there were old ones * refactor: Unify viewport transfer: use common variable names * fix: Use same saved cursor logic in height resize as width See zellij-org#2182 for original introduction that only added it in one branch, this fixes an issue where the saved cursor was incorrectly reset when the real cursor was * fix: Correct saved+real cursor calculations when reflowing long lines * fix: Don't create canonical lines if cursor ends on EOL after resize Previously if a 20 character line were split into two 10 character lines, the cursor would be placed on the line after the two lines. New characters would then be treated as a new canonical line. This commit fixes this by biasing cursors to the end of the previous line. * fix: for cursor index calculation in lines that are already wrapped * chore: test for real/saved cursor position being handled separately * chore: Apply cargo format * chore(repo): update issue templates * Bump rust version to 1.75.0 (zellij-org#3039) * rust-toolchain: Bump toolchain version to 1.69.0 which, compared to the previous 1.67.0, has the following impacts on `zellij`: - [Turn off debuginfo for build deps][2]: Increases build time (on my machine) from ~230 s in 1.67.0 to ~250 s now, *which is unexpected* This version also changes [handling of the `default-features` flag][3] when specifying dependencies in `Cargo.toml`. If a dependent crate requires `default-features = true` on a crate that is required as `default-features = false` further up the dependency tree, the `true` setting "wins". We only specify `default-features = false` for three crates total: - `names`: This is used only by us - `surf`: This is used only by us - `vte`: This is also required by `strip-ansi-escapes`, but that has `default-features = false` as well How this affects our transitive dependencies is unknown at this point. [2]: rust-lang/cargo#11252 [3]: rust-lang/cargo#11409 * rust-toolchain: Bump toolchain version to 1.70.0 which, compared to the previous 1.69.0, as the following impacts on `zellij`: 1. [Enable sparse registry checkout for crates.io by default][1] This drastically increases the time to first build on a fresh rust installation/a rust installation with a clean cargo registry cache. Previously it took about 75s to populate the deps/cache (with `cargo fetch --locked` and ~100 MBit/s network), whereas now the same process takes ~10 s. 2. [The `OnceCell` type is now part of std][2] In theory, this would allow us to cut a dependency from `zellij-utils`, but the `once_cell` crate is pulled in by another 16 deps, so there's no point in attempting it right now. Build times and binary sizes are unaffected by this change compared to the previous 1.69.0 toolchain. [1]: rust-lang/cargo#11791 [2]: https://doc.rust-lang.org/stable/std/cell/struct.OnceCell.html * rust-toolchain: Bump toolchain version to 1.75.0 which, compared to the previous 1.70.0, has the following impacts on `zellij`: 1. [cross-crate inlining][8] This should increase application performance, as functions can now be inlined across crates. 2. [`async fn` in traits][9] This would allow us to drop the `async_trait` dependency, but it is currently still required by 3 other dependencies. Build time in debug mode (on my own PC) is cut down from 256s to 189s (for a clean build). Build time in release mode is cut down from 473s to 391s (for a clean build). Binary sizes only change minimally (825 MB -> 807 MB in debug, 29 MB -> 30 MB in release). [8]: rust-lang/rust#116505 [9]: rust-lang/rust#115822 * chore: Apply rustfmt. * CHANGELOG: Add PR zellij-org#3039. * feat(plugins): introduce 'pipes', allowing users to pipe data to and control plugins from the command line (zellij-org#3066) * prototype - working with message from the cli * prototype - pipe from the CLI to plugins * prototype - pipe from the CLI to plugins and back again * prototype - working with better cli interface * prototype - working after removing unused stuff * prototype - working with launching plugin if it is not launched, also fixed event ordering * refactor: change message to cli-message * prototype - allow plugins to send messages to each other * fix: allow cli messages to send plugin parameters (and implement backpressure) * fix: use input_pipe_id to identify cli pipes instead of their message name * fix: come cleanups and add skip_cache parameter * fix: pipe/client-server communication robustness * fix: leaking messages between plugins while loading * feat: allow plugins to specify how a new plugin instance is launched when sending messages * fix: add permissions * refactor: adjust cli api * fix: improve cli plugin loading error messages * docs: cli pipe * fix: take plugin configuration into account when messaging between plugins * refactor: pipe message protobuf interface * refactor: update(event) -> pipe * refactor - rename CliMessage to CliPipe * fix: add is_private to pipes and change some naming * refactor - cli client * refactor: various cleanups * style(fmt): rustfmt * fix(pipes): backpressure across multiple plugins * style: some cleanups * style(fmt): rustfmt * style: fix merge conflict mistake * style(wording): clarify pipe permission * docs(changelog): introduce pipes * fix: add some robustness and future proofing * fix e2e tests --------- Co-authored-by: Aram Drevekenin <[email protected]> Co-authored-by: har7an <[email protected]> * fix integer overflow again (oops) --------- Co-authored-by: Aram Drevekenin <[email protected]> Co-authored-by: har7an <[email protected]>
Pkgsrc changes: * Adjust patches and cargo checksums to new versions. * For an external LLVM, set dependency of llvm >= 16, in accordance with the upstream changes. * Mark that on NetBSD we now need >= 9.0, so 8.x is no longer supported. * On NetBSD/sparc64 10.x, we now need GCC 12 to build the embedded LLVM, which is version 17; apparently GCC 10.4 or 10.5 mis-compiles it, resulting in an illegal instruction fault during the build. Ref. rust-lang/rust#117231 Upstream changes: Version 1.75.0 (2023-12-28) ========================== - [Stabilize `async fn` and return-position `impl Trait` in traits.] (rust-lang/rust#115822) - [Allow function pointer signatures containing `&mut T` in `const` contexts.] (rust-lang/rust#116015) - [Match `usize`/`isize` exhaustively with half-open ranges.] (rust-lang/rust#116692) - [Guarantee that `char` has the same size and alignment as `u32`.] (rust-lang/rust#116894) - [Document that the null pointer has the 0 address.] (rust-lang/rust#116988) - [Allow partially moved values in `match`.] (rust-lang/rust#103208) - [Add notes about non-compliant FP behavior on 32bit x86 targets.] (rust-lang/rust#113053) - [Stabilize ratified RISC-V target features.] (rust-lang/rust#116485) Compiler -------- - [Rework negative coherence to properly consider impls that only partly overlap.] (rust-lang/rust#112875) - [Bump `COINDUCTIVE_OVERLAP_IN_COHERENCE` to deny, and warn in dependencies.] (rust-lang/rust#116493) - [Consider alias bounds when computing liveness in NLL.] (rust-lang/rust#116733) - [Add the V (vector) extension to the `riscv64-linux-android` target spec.] (rust-lang/rust#116618) - [Automatically enable cross-crate inlining for small functions] (rust-lang/rust#116505) - Add several new tier 3 targets: - [`csky-unknown-linux-gnuabiv2hf`] (rust-lang/rust#117049) - [`i586-unknown-netbsd`] (rust-lang/rust#117170) - [`mipsel-unknown-netbsd`] (rust-lang/rust#117356) Refer to Rust's [platform support page][platform-support-doc] for more information on Rust's tiered platform support. Libraries --------- - [Override `Waker::clone_from` to avoid cloning `Waker`s unnecessarily.] (rust-lang/rust#96979) - [Implement `BufRead` for `VecDeque<u8>`.] (rust-lang/rust#110604) - [Implement `FusedIterator` for `DecodeUtf16` when the inner iterator does.] (rust-lang/rust#110729) - [Implement `Not, Bit{And,Or}{,Assign}` for IP addresses.] (rust-lang/rust#113747) - [Implement `Default` for `ExitCode`.] (rust-lang/rust#114589) - [Guarantee representation of None in NPO] (rust-lang/rust#115333) - [Document when atomic loads are guaranteed read-only.] (rust-lang/rust#115577) - [Broaden the consequences of recursive TLS initialization.] (rust-lang/rust#116172) - [Windows: Support sub-millisecond sleep.] (rust-lang/rust#116461) - [Fix generic bound of `str::SplitInclusive`'s `DoubleEndedIterator` impl] (rust-lang/rust#100806) - [Fix exit status / wait status on non-Unix `cfg(unix)` platforms.] (rust-lang/rust#115108) Stabilized APIs --------------- - [`Atomic*::from_ptr`] (https://doc.rust-lang.org/stable/core/sync/atomic/struct.AtomicUsize.html#method.from_ptr) - [`FileTimes`] (https://doc.rust-lang.org/stable/std/fs/struct.FileTimes.html) - [`FileTimesExt`] (https://doc.rust-lang.org/stable/std/os/windows/fs/trait.FileTimesExt.html) - [`File::set_modified`] (https://doc.rust-lang.org/stable/std/fs/struct.File.html#method.set_modified) - [`File::set_times`] (https://doc.rust-lang.org/stable/std/fs/struct.File.html#method.set_times) - [`IpAddr::to_canonical`] (https://doc.rust-lang.org/stable/core/net/enum.IpAddr.html#method.to_canonical) - [`Ipv6Addr::to_canonical`] (https://doc.rust-lang.org/stable/core/net/struct.Ipv6Addr.html#method.to_canonical) - [`Option::as_slice`] (https://doc.rust-lang.org/stable/core/option/enum.Option.html#method.as_slice) - [`Option::as_mut_slice`] (https://doc.rust-lang.org/stable/core/option/enum.Option.html#method.as_mut_slice) - [`pointer::byte_add`] (https://doc.rust-lang.org/stable/core/primitive.pointer.html#method.byte_add) - [`pointer::byte_offset`] (https://doc.rust-lang.org/stable/core/primitive.pointer.html#method.byte_offset) - [`pointer::byte_offset_from`] (https://doc.rust-lang.org/stable/core/primitive.pointer.html#method.byte_offset_from) - [`pointer::byte_sub`] (https://doc.rust-lang.org/stable/core/primitive.pointer.html#method.byte_sub) - [`pointer::wrapping_byte_add`] (https://doc.rust-lang.org/stable/core/primitive.pointer.html#method.wrapping_byte_add) - [`pointer::wrapping_byte_offset`] (https://doc.rust-lang.org/stable/core/primitive.pointer.html#method.wrapping_byte_offset) - [`pointer::wrapping_byte_sub`] (https://doc.rust-lang.org/stable/core/primitive.pointer.html#method.wrapping_byte_sub) These APIs are now stable in const contexts: - [`Ipv6Addr::to_ipv4_mapped`] (https://doc.rust-lang.org/stable/core/net/struct.Ipv6Addr.html#method.to_ipv4_mapped) - [`MaybeUninit::assume_init_read`] (https://doc.rust-lang.org/stable/core/mem/union.MaybeUninit.html#method.assume_init_read) - [`MaybeUninit::zeroed`] (https://doc.rust-lang.org/stable/core/mem/union.MaybeUninit.html#method.zeroed) - [`mem::discriminant`] (https://doc.rust-lang.org/stable/core/mem/fn.discriminant.html) - [`mem::zeroed`] (https://doc.rust-lang.org/stable/core/mem/fn.zeroed.html) Cargo ----- - [Add new packages to `[workspace.members]` automatically.] (rust-lang/cargo#12779) - [Allow version-less `Cargo.toml` manifests.] (rust-lang/cargo#12786) - [Make browser links out of HTML file paths.] (rust-lang/cargo#12889) Rustdoc ------- - [Accept less invalid Rust in rustdoc.] (rust-lang/rust#117450) - [Document lack of object safety on affected traits.] (rust-lang/rust#113241) - [Hide `#[repr(transparent)]` if it isn't part of the public ABI.] (rust-lang/rust#115439) - [Show enum discriminant if it is a C-like variant.] (rust-lang/rust#116142) Compatibility Notes ------------------- - [FreeBSD targets now require at least version 12.] (rust-lang/rust#114521) - [Formally demote tier 2 MIPS targets to tier 3.] (rust-lang/rust#115238) - [Make misalignment a hard error in `const` contexts.] (rust-lang/rust#115524) - [Fix detecting references to packed unsized fields.] (rust-lang/rust#115583) - [Remove support for compiler plugins.] (rust-lang/rust#116412)
This is basically reviving #70550
The
#[inline]
attribute can have a significant impact on code generation or runtime performance (because it enables inlining between CGUs where it would normally not happen) and also on compile-time performance (because it enables MIR inlining). But it has to be added manually, which is awkward.This PR factors whether a DefId is cross-crate inlinable into a query, and replaces all uses of
CodegenFnAttrs::requests_inline
with this new query. The new query incorporates all the other logic that is used to determine whether a Def should be treated as cross-crate-inlinable, and as a last step inspects the function's optimized_mir to determine if it should be treated as cross-crate-inlinable.The heuristic implemented here is deliberately conservative; we only infer inlinability for functions whose optimized_mir does not contain any calls or asserts. I plan to study adjusting the cost model later, but for now the compile time implications of this change are so significant that I think this very crude heuristic is well worth landing.