Passing ziptime-info=false to ZipFileSystemProvider (along with other classpath performance ideas)

[Jason Zaugg]

Submitted as https://github.com/openjdk/jdk/pull/24176

Jason Zaugg

On Sat, 22 Mar 2025 at 20:06, Jaikiran Pai <jai.forums2013 at gmail.com> wrote:

> Hello Jason,
>
> For the proposal to set zipinfo-time=false when constructing the zipfs
> FileSystem in the compiler code, I've created
> https://bugs.openjdk.org/browse/JDK-8352642. Since you have already
> experimented with the changes, could you create a PR with just that change
> for review?
>
> -Jaikiran
> On 17/03/25 10:15 am, Jason Zaugg wrote:
>
> On Sun, 16 Mar 2025 at 19:40, Jaikiran Pai <jai.forums2013 at gmail.com>
> wrote:
>
>> Hello Jason,
>>
>> On 02/03/25 7:13 pm, Jason Zaugg wrote:
>> > I've been analysing the performance of javac in a relatively large
>> > codebase and found classpath scanning to be the dominant factor. There
>> > are large numbers of compilation tasks, most with long classpaths
>> > including large JARs.
>> >
>> > I've been using the wall-time profiling mode of async-profiler to
>> > analyse where time is being spent.
>> >
>> > I've been able improve build times (most dramatically on Windows, 750s
>> > reduced to 350s) by modifying javac to:
>> >
>> > 1. pass ziptime-info=false to the env Map of
>> > jarFsProvider.newFileSystem, which means the classpath scan in the
>> > constructor of ArchiveContainer need only read the ZIP central
>> > directory, rather then reading per-entry Extended Time metadata.
>>
>> I am not familiar with the compiler implementation code, but speaking in
>> context of the ZipFileSystem, it appears that this (undocumented)
>> "ziptime-info" property of the ZIP file system was indeed introduced to
>> help improve the performance of the ZipFileSystem. Setting it to false
>> implies that the lastAccessTime() and the creationTime() returned by the
>> ZipFileAttributes may not be accurate for some entries, since settings
>> this property to false skips reading the per entry creation and access
>> time information for the entries. Would you happen to know if either of
>> these two methods are used in the code path where you are proposing to
>> add "ziptime-info=false"?
>>
>> -Jaikiran
>>
>
> I am not aware of any use of these attributes within javac, nor does it
> seem likely that they would be used.
>
> But the difference in the attributes could be observed by a client of
> JavaFileManager that
> unwraps the underlying Path from a JavaFileObject representing a JAR entry.
>
> ```
> jshell> var compiler = javax.tools.ToolProvider.getSystemJavaCompiler();
> compiler ==> com.sun.tools.javac.api.JavacTool at 4b9e255
>
> jshell> var fileManager = compiler.getStandardFileManager(null, null, null)
> fileManager ==> com.sun.tools.javac.file.JavacFileManager at 7f416310
>
> jshell> var head =
> fileManager.list(javax.tools.StandardLocation.CLASS_PATH, "",
> Set.of(javax.tools.JavaFileObject.Kind.OTHER), true).iterator().next()
> head ==> JarFileObject[/Users/jz/Library/Caches/Coursier/v ...
> e/protobuf/wrappers.proto]
>
> jshell> java.nio.file.Files.readAttributes(fileManager.asPath(head),
> java.nio.file.attribute.BasicFileAttributes.class)
> $11 ==>     name            : /google/protobuf/wrappers.proto
>     creationTime    : Fri Feb 14 21:06:54 AEST 2020
>     lastAccessTime  : Fri Feb 14 21:06:54 AEST 2020
>     lastModifiedTime: Fri Feb 14 21:06:54 AEST 2020
> ...
> ```
>
> A new option could be added to toggle between the existing and proposed
> behaviour if this is a concern.
>
> Regards,
>
> Jason Zaugg
>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://mail.openjdk.org/pipermail/compiler-dev/attachments/20250323/f9324a08/attachment.htm>