Purity inference #7170

agu-z · 2024-10-17T00:32:01Z

Introduces the "Purity Inference" model of effects as described in Richard's talk:

All functions are now either pure or effectful

pure : Str -> Str
effectful! : Str => Str

Functions are inferred as effectful when they call other effectful functions

printName! = \user ->
    putLine! user.name

The "source" effect functions come from the platform.
Higher-order functions that are annotated to take pure functions cannot be passed effectful functions

List.map names putLine! 
             # ^^^^^^^^ error

However, you can make an effectful higher-order function that takes effectful functions

forEach! : List a, (a => {}) => {}
forEach! = \list, fn! -> ...

A warning is produced if effectful functions are not suffixed with ! and viceversa
The ! suffix is not only required on defs, but also on record fields, and all pattern matched identifiers (tuple, tags, opaques, etc).
Standalone statements are allowed in effectful functions:

main! = \{} ->
    line = getLine! {}
    
    if line == "secret" then
        putLine! "You found the secret"
        putLine! "Congratulations!"
    else
        {}
    
    putLine! "You entered: $(line)"

A warning is produced if they don't call any effectful functions, since that means they are dead code
Also, an IGNORED RESULT warning is produced if they return anything other than {}
If you still want to ignore the result, you can assign it to _:

_ = File.delete! "data.json"

...but you'll still get a warning if what you ignored is pure
You can roc run a program with a pure/effectful mismatch, which can be useful while you're debugging
Task still works if that's what the platform exposes in its hosted module. This will allow us to migrate platforms incrementally.

Type checking and errors preview:
https://www.youtube.com/watch?v=5a3r88-oEQQ

ayazhafiz · 2024-10-17T00:37:16Z

awesome timing

…

smores56

Initial review, only partway through. More comments tomorrow!

crates/compiler/can/src/desugar.rs

crates/compiler/parse/src/ident.rs

crates/compiler/problem/src/can.rs

crates/compiler/can/src/constraint.rs

crates/compiler/can/src/copy.rs

crates/compiler/can/src/def.rs

crates/compiler/can/tests/snapshots/test_suffixed__suffixed_tests__apply_argument_multiple.snap

crates/compiler/can/src/expr.rs

crates/compiler/mono/src/layout.rs

examples/cli/effects.roc

smores56

Okay, I've finished reading the whole thing (more or less). It looks good to me! A couple of minor things, and then I think we're good to go. You supporting both Tasks and effectful functions makes this need much less coordination to get right than the built-in Task effort.

crates/compiler/module/src/symbol.rs

crates/reporting/src/error/type.rs

crates/compiler/load/tests/test_reporting.rs

ayazhafiz · 2024-10-29T19:24:00Z

@agu-z can you ping me for review after rebase?

ayazhafiz · 2024-10-31T01:32:01Z

crates/compiler/can/src/constraint.rs

+    /// Require idents to be accurately suffixed
+    FxSuffix(Index<FxSuffixConstraint>),
+    /// Set an fx var as pure if flex (no effectful functions were called)
+    FlexToPure(Variable),


Is this necessary? Would it be sufficient to add an Eq constraint that the fx var on a function is Pure or Impure depending on the suffix? With an appropriate context.

Do you mean FlexToPure or FxSuffix should use Eq?

The former I think wouldn't work because the logic is different than the normal unification of fx vars. However, FxSuffix might work as Eq with a custom Reason.

@ayazhafiz let me know which one you mean when you have a chance

We decided to merge this, but I'm happy to clean this up in a follow up PR!

This is because the NoIdentifiersIntroduced error was moved to the type checker.

Non-top-level defs are already covered

Things go wrong when multiple tests depend on the same platform on Linux. Tried a few workarounds but the problem persisted. We decided to keep only one test for now.

lukewilliamboswell

Massive upgrade, thank you for all the hard work Agus 🎉 😍

agu-z force-pushed the purity-inference branch 8 times, most recently from 9039954 to c063025 Compare October 25, 2024 00:03

agu-z changed the title ~~[WIP] Purity inference~~ Purity inference Oct 25, 2024

agu-z marked this pull request as ready for review October 25, 2024 00:19

agu-z requested a review from ayazhafiz October 25, 2024 00:19

agu-z force-pushed the purity-inference branch from c063025 to ee94d81 Compare October 25, 2024 00:22

smores56 reviewed Oct 27, 2024

View reviewed changes

crates/compiler/mono/src/layout.rs Outdated Show resolved Hide resolved

crates/compiler/mono/src/layout.rs Outdated Show resolved Hide resolved

examples/cli/effects.roc Show resolved Hide resolved

lukewilliamboswell mentioned this pull request Oct 28, 2024

Upgrade to purity inference lukewilliamboswell/roc-ray#38

Merged

smores56 reviewed Oct 28, 2024

View reviewed changes

agu-z force-pushed the purity-inference branch 2 times, most recently from dcbff1f to 2b8d9aa Compare October 30, 2024 23:27

ayazhafiz reviewed Oct 31, 2024

View reviewed changes

agu-z force-pushed the purity-inference branch 4 times, most recently from a5aeb82 to d3abe4a Compare November 7, 2024 05:40

agu-z added 5 commits November 7, 2024 18:54

Parse effectful arrow in function annotations

ef4eeb5

Format effectful arrows in annotations

d692fc7

Add effect_type to can ClosureData

386a505

Rename effect_type to fx_type and add to FunctionDef

75177c9

Add fx var to Type::Function et al

3cef756

agu-z added 23 commits November 7, 2024 18:54

Treat untyped unsuffixed functions as pure

af6fc63

Treat untyped suffixed functions as effectful

cfc4be5

Add hint about forgetting to call a function

175a2b5

Allow ignored defs with an effectful RHS

c9f001b

Use byte literal instead of cast and ignore too_many_args

a2f940b

Expect only one problem in test_can::shadow_annotation

5f5e123

This is because the NoIdentifiersIntroduced error was moved to the type checker.

update mono tests: ids increase because of new fx vars

ca7697d

Add simple effectful cli run tests

de2260e

update ui test

be0afbc

Make sure to drop suffix from symbols exposed to the host

e3c6b75

Explicit message for StmtAfterExpr in desugar

2e5c143

Print fx_suffix_constraints in Debug impl for Constraints

b31b30c

Do not alias ClosureData.fx_type in pattern matches

dae10d1

Refactor if-let to let-else-continue

46e808d

Update region when desugaring ! in Task mode

a0f4b38

Return early when encountering ! in an ident

1a3d8ce

Do not error when encountering EffectfulFunc in lambda_set_size

5848a3e

Add explicit error for EffectfulFunc in layout

2fdd1ef

Use plural 'effects' in FxInTopLevel error

935d460

Fix 'right-hand side' typo in errors

1e835bb

Improve wording in pattern suffix errors

a4296ca

Restore UNNCESSARY DEFINITION errors for top-level defs

bc0cfef

Non-top-level defs are already covered

Update regions in suffixed tests

5641884

agu-z force-pushed the purity-inference branch from 5c2edf6 to a4ccef9 Compare November 7, 2024 21:54

Ignore all but one cli_run effectful tests on Linux

7f8149d

Things go wrong when multiple tests depend on the same platform on Linux. Tried a few workarounds but the problem persisted. We decided to keep only one test for now.

agu-z force-pushed the purity-inference branch from a4ccef9 to 7f8149d Compare November 7, 2024 22:12

lukewilliamboswell approved these changes Nov 7, 2024

View reviewed changes

agu-z merged commit dfb2966 into main Nov 7, 2024
18 checks passed

agu-z deleted the purity-inference branch November 8, 2024 00:00

smores56 mentioned this pull request Nov 12, 2024

Upgrade to use purity inference smores56/weaver#6

Open

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Purity inference #7170

Purity inference #7170

agu-z commented Oct 17, 2024 •

edited

Loading

ayazhafiz commented Oct 17, 2024 via email

smores56 left a comment

smores56 left a comment

ayazhafiz commented Oct 29, 2024

ayazhafiz Oct 31, 2024

agu-z Oct 31, 2024

agu-z Nov 5, 2024

agu-z Nov 8, 2024

lukewilliamboswell left a comment

Purity inference #7170

Purity inference #7170

Conversation

agu-z commented Oct 17, 2024 • edited Loading

ayazhafiz commented Oct 17, 2024 via email

smores56 left a comment

Choose a reason for hiding this comment

smores56 left a comment

Choose a reason for hiding this comment

ayazhafiz commented Oct 29, 2024

ayazhafiz Oct 31, 2024

Choose a reason for hiding this comment

agu-z Oct 31, 2024

Choose a reason for hiding this comment

agu-z Nov 5, 2024

Choose a reason for hiding this comment

agu-z Nov 8, 2024

Choose a reason for hiding this comment

lukewilliamboswell left a comment

Choose a reason for hiding this comment

agu-z commented Oct 17, 2024 •

edited

Loading