Run processes without forced GC or a login shell - #32
Open
lm-sousa wants to merge 3 commits into
Open
Conversation
Introduced in 530a468 as an experiment to reduce fork memory. Measured at 50s of GC pauses in a 98s test run vs 0.14s with automatic GC only. Automatic GC still runs; removing this does not disable collection.
Drop the bash -l -c wrapper added in 44d1d93. The wrapper added ~50ms per launch for shell parsing, broke argument fidelity (naive space escaping), and re-sourced login profiles over the inherited environment instead of carrying it faithfully. ProcessBuilder children already inherit the JVM's full environment, user, and working directory. All in-tree callers pass plain argv; shell semantics for string commands are no longer supported on Linux.
lm-sousa
force-pushed
the
clava-optimizations
branch
3 times, most recently
from
September 5, 2026 01:23
da92ff7 to
21dc912
Compare
With direct argv, a missing or non-executable command now throws from
runProcess(), as its javadoc always documented ('If there is any
problem with the process, throws an exception') - the removed shell
wrapper used to mask launch failures as exit code 127. Handle the
launch failure where graceful failure is the intent:
- SpecsGraphviz: dot being absent is a normal condition, so checkDot()
returns false and renderDot() logs instead of crashing.
- ProcessExecution: the jobs framework is built on return codes with
no exception handling, so run() reports the failed launch as
exit code 127 instead of propagating.
This also restores SpecsSystem.isCommandAvailable()'s intended
behavior: it detects missing commands by catching the launch
exception, which the wrapper's masking previously defeated.
lm-sousa
force-pushed
the
clava-optimizations
branch
from
September 5, 2026 01:37
21dc912 to
44ab197
Compare
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Every
SpecsSystem.runProcess()call paid ~60 ms of overhead: a forcedSystem.gc()before every fork (added as an experiment in 2018 to shrink fork memory) and, on Linux, a rewrap of the argv intobash -l -c(added in 2021). The wrapper also re-sourced login profiles on top of the inherited environment and broke argument fidelity with a naive space-escaping hack.This removes both.
ProcessBuilderchildren already inherit the JVM's full environment, user, and working directory at fork/exec — strictly closer to the parent than the wrapper, which a probe showed mutating PATH (duplicate entry from profile re-sourcing). Automatic garbage collection still runs; only the forced full collection was removed. The Windowscmd /cbranch is untouched, and all in-tree callers pass plain argv.Measured impact (Clava-JS test suite, 155 passed | 9 skipped in every run, six measured runs): median 111.79 s → 47.60 s (−64.19 s, 57.4%); Welch two-sample t-test vs the pre-change distribution: t(6.28) = 16.51, p = 0.0000021, 95% CI [51.28, 68.90] s. Per-launch probe: 60.29 ms → 7.65 ms (
echolaunch, 200 iterations). The suite performs on the order of a thousand process launches per run.Also removes the GC call's per-launch measurement logging introduced with the same experiment.