fix: Deflater dictionary data loss, PATH lookup and other review findings, and narrower PR test runs by Project516 · Pull Request #72 · Project516/java-web

Refresh test/jtreg/times.tsv from a full CI run so shards balance by CI's own test times

Project516

committed

5068394

Select the OpenJDK test directories a changed file can affect, plus a sample of the rest.

Project516

committed

adff6ea

docs: follow Conventional Commits and add a pull request template

Project516

committed

5afe92c

fix(deflate): keep pending symbols when a raw dictionary is set mid-stream

Project516

committed

02a2b75

test(jit): cover receiver guard misses at a top-level call and inside an inlined region

Project516

committed

1599cab

Search PATH the way the JDK does, with its ELOOP rule and exec failure wording.

Project516

committed

dab97a1

Report an unlimited open file limit as -1, as HotSpot does.

Project516

committed

94566f2

Throw an OutOfMemoryError even when the hook that reports it fails.

Project516

committed

8587e44

Run OutOfMemoryError commands even when their output cannot be written.

Project516

committed

6d24d2c

Treat an unreadable host entry as unknown, so watching it cannot throw or hide a change.

Project516

committed

0a00400

Encode the child environment when it holds a variable named JAVA_WEB_ENV.

Project516

committed

d5e86f9

Keep every direct buffer reachable in DirectMemoryLimit so the GC cannot change the output.

Project516

committed

029d2de

Expand every jtreg test property in @run lines, including the main class.

Project516

committed

8eb887f

docs: say that a SecureDirectoryStream follows its directory's path

Project516

committed

f8dcadd

Reuse an open file's bytes only for the same inode, including files streamed from the host.

Project516

committed

191ee79

Keep an open successful when its inode pin cannot be taken, and test it.

Project516

committed

c0dd40d

ci(jtreg): select URLClassLoader tests for zip changes and tests that leave locale-data.txt

Project516

committed

0d09383

test: keep the open file limit program out of the generic HotSpot comparison

Project516

committed

0a75ecf

test: make the symlink and receiver guard programs tell their cases apart

Project516

committed

dd7285c

test: lower the open file limit with prlimit instead of a shell

Project516

committed

92676ef

Navigation Menu

Repository navigation

fix: Deflater dictionary data loss, PATH lookup and other review findings, and narrower PR test runs - #72

fix: Deflater dictionary data loss, PATH lookup and other review findings, and narrower PR test runs#72
Project516 merged 20 commits into
masterfrom
p516/ci-speed

Commits on Nov 7, 2026

Refresh test/jtreg/times.tsv from a full CI run so shards balance by CI's own test times

Select the OpenJDK test directories a changed file can affect, plus a sample of the rest.

docs: follow Conventional Commits and add a pull request template

fix(deflate): keep pending symbols when a raw dictionary is set mid-stream

test(jit): cover receiver guard misses at a top-level call and inside an inlined region

Search PATH the way the JDK does, with its ELOOP rule and exec failure wording.

Report an unlimited open file limit as -1, as HotSpot does.

Throw an OutOfMemoryError even when the hook that reports it fails.

Run OutOfMemoryError commands even when their output cannot be written.

Treat an unreadable host entry as unknown, so watching it cannot throw or hide a change.

Encode the child environment when it holds a variable named JAVA_WEB_ENV.

Keep every direct buffer reachable in DirectMemoryLimit so the GC cannot change the output.

Expand every jtreg test property in @run lines, including the main class.

docs: say that a SecureDirectoryStream follows its directory's path

Reuse an open file's bytes only for the same inode, including files streamed from the host.

Keep an open successful when its inode pin cannot be taken, and test it.

ci(jtreg): select URLClassLoader tests for zip changes and tests that leave locale-data.txt

test: keep the open file limit program out of the generic HotSpot comparison

test: make the symlink and receiver guard programs tell their cases apart

test: lower the open file limit with prlimit instead of a shell