-
Notifications
You must be signed in to change notification settings - Fork 4.1k
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
[8.0.0] Fix most Unicode encoding bugs #24260
Conversation
Bazel aims to support arbitrary file system path encodings (even raw byte sequences) by attempting to force the JVM to use a Latin-1 locale for OS interactions. As a result, Bazel internally encodes `String`s as raw byte arrays with a Latin-1 coder and no encoding information. Whenever it interacts with encoding-aware APIs, this may require a reencoding of the `String` contents, depending on the OS and availability of a Latin-1 locale. This PR introduces the concepts of *internal*, *Unicode*, and *platform* strings and adds dedicated optimized functions for converting between these three types (see the class comment on the new `StringEncoding` helper class for details). These functions are then used to standardize and fix conversion throughout the code base. As a result, a number of new end-to-end integration tests for the handling of Unicode in file paths, command-line arguments and environment variables now pass. Full support for Unicode beyond the current active code page on Windows is left to a follow-up PR as it may require patching the embedded JDK. * Replace ad-hoc conversion logic with the new consistent set of helper functions. * Make more parts of the Bazel client's Windows implementation Unicode-aware. This also fixes the behavior of `SetEnv` on Windows, which previously would remove an environment variable if passed an empty value for it, which doesn't match the Unix behavior. * Drop the `charset` parameter from all methods related to parameter files. The `ISO-8859-1` vs. `UTF-8` choice was flawed since Bazel's internal string representation doesn't maintain any encoding information - `ISO-8859-1` just meant "write out raw bytes", which is the only choice that matches what arguments would look like if passed on the command line. * Convert server args to the internal string representation. The arguments for requests to the server were already converted to Bazel's internal string representation, which resulted in a mismatch between `--client_cwd` and `--workspace_directory` if the workspace path contains non-ASCII characters. * Read the downloader config using Bazel's filesystem implementation. * Make `MacOSXFsEventsDiffAwareness` UTF-8 aware. It previously used the `GetStringUTF` JNI method, which, despite its name, doesn't return the UTF-8 representation of a string, but modified CESU-8 (nobody ever wants this). * Correctly reencode path strings for `LocalDiffAwareness`. * Correctly reencode the value of `user.dir`. * Correctly turn `ExecRequest` fields into strings for `ProcessBuilder` for `bazel --batch run`. This makes it possible to reenable the `test_consistent_command_line_encoding` test, fixing bazelbuild#1775. * Fix encoding issues in `TargetCompleteEvents`. * Fix encoding issues in `SubprocessFactory` implementations. * Drop obsolete warning if `file.encoding` doesn't equal `ISO-8859-1` as file names are encoded with `sun.jnu.encoding` now. * Consistently reencode internal strings passed into and out of `FileSystem` implementations, e.g. if reading a symlink target. Tests are added that verify the interaction between `FileSystem` implementations and the Java (N)IO APIs on Unicode file paths. Fixes bazelbuild#1775. Fixes bazelbuild#11602. Fixes bazelbuild#18293. Work towards #374. Work towards bazelbuild#23859. Closes bazelbuild#24010. PiperOrigin-RevId: 694114597 Change-Id: I5bdcbc14a90dd1f0f34698aebcbd07cd2bde7a23
FYI @fmeum please take a look at the test failures |
@iancha1992 This patch should fix the build, could you apply it? diff --git a/src/main/java/com/google/devtools/build/lib/BUILD b/src/main/java/com/google/devtools/build/lib/BUILD
index 71deae4ac9..898ef5a81d 100644
--- a/src/main/java/com/google/devtools/build/lib/BUILD
+++ b/src/main/java/com/google/devtools/build/lib/BUILD
@@ -325,6 +325,7 @@ java_library(
"//src/main/java/com/google/devtools/build/lib/actions:file_metadata",
"//src/main/java/com/google/devtools/build/lib/actions:important_output_handler",
"//src/main/java/com/google/devtools/build/lib/actions:localhost_capacity",
+ "//src/main/java/com/google/devtools/build/lib/actions:middleman_type",
"//src/main/java/com/google/devtools/build/lib/actions:package_roots",
"//src/main/java/com/google/devtools/build/lib/actions:resource_manager",
"//src/main/java/com/google/devtools/build/lib/actions:runfiles_tree", |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
The original (head) PR is getting rolled back - please don't submit this cherry pick until we figure out what to do
|
Bazel aims to support arbitrary file system path encodings (even raw byte sequences) by attempting to force the JVM to use a Latin-1 locale for OS interactions. As a result, Bazel internally encodes
String
s as raw byte arrays with a Latin-1 coder and no encoding information. Whenever it interacts with encoding-aware APIs, this may require a reencoding of theString
contents, depending on the OS and availability of a Latin-1 locale.This PR introduces the concepts of internal, Unicode, and platform strings and adds dedicated optimized functions for converting between these three types (see the class comment on the new
StringEncoding
helper class for details). These functions are then used to standardize and fix conversion throughout the code base. As a result, a number of new end-to-end integration tests for the handling of Unicode in file paths, command-line arguments and environment variables now pass.Full support for Unicode beyond the current active code page on Windows is left to a follow-up PR as it may require patching the embedded JDK.
SetEnv
on Windows, which previously would remove an environment variable if passed an empty value for it, which doesn't match the Unix behavior.charset
parameter from all methods related to parameter files. TheISO-8859-1
vs.UTF-8
choice was flawed since Bazel's internal string representation doesn't maintain any encoding information -ISO-8859-1
just meant "write out raw bytes", which is the only choice that matches what arguments would look like if passed on the command line.--client_cwd
and--workspace_directory
if the workspace path contains non-ASCII characters.MacOSXFsEventsDiffAwareness
UTF-8 aware. It previously used theGetStringUTF
JNI method, which, despite its name, doesn't return the UTF-8 representation of a string, but modified CESU-8 (nobody ever wants this).LocalDiffAwareness
.user.dir
.ExecRequest
fields into strings forProcessBuilder
forbazel --batch run
. This makes it possible to reenable thetest_consistent_command_line_encoding
test, fixing Fix (and reenable) run_test.test_consistent_command_line_encoding #1775.TargetCompleteEvents
.SubprocessFactory
implementations.file.encoding
doesn't equalISO-8859-1
as file names are encoded withsun.jnu.encoding
now.FileSystem
implementations, e.g. if reading a symlink target. Tests are added that verify the interaction betweenFileSystem
implementations and the Java (N)IO APIs on Unicode file paths.Fixes #1775.
Fixes #11602.
Fixes #18293.
Work towards #374.
Work towards #23859.
Closes #24010.
PiperOrigin-RevId: 694114597
Change-Id: I5bdcbc14a90dd1f0f34698aebcbd07cd2bde7a23
Commit a58fe3f