Bug #14457: --jit outputs error when script changes $TMPDIR

• Status: Closed
• Assignee: k0kubun (Takashi Kokubun)
• Target version: 2.6
• ruby -v: ruby 2.6.0dev (2018-02-07 trunk 62260) [x86_64-linux]
• Backport: 2.3: DONTNEED, 2.4: DONTNEED, 2.5: DONTNEED

[k0kubun (Takashi Kokubun)]

Shibata-san reported the following output when running bundler's test with --jit (it looks implicitly passed).

~/D/g/b/bundler (master) > rspec spec/resolver/basic_spec.rb 
Run options: exclude {:bundler=>"=< 2.0", :rubygems_master=>true, :git=>"=< 2.16.1", :rubygems=>"=< 2.7.3", :ruby=>"=< 2.6.0", :realworld=>true, :sudo=>true}

Resolving
error: unable to open output file '/Users/hsbt/Documents/github.com/bundler/bundler/tmp/tmpdir/_ruby_mjitp36578u588-067cfe.o':
      'No such file or directory'
1 error generated.
  resolves a single gem
  resolves a gem with dependencies
  resolves a conflicting index
  resolves a complex conflicting index
  resolves a index with conflict on child
  prefers explicitly requested dependencies when resolving an index which would otherwise be ambiguous
  prefers non-prerelease resolutions in sort order
  resolves a index with root level conflict on child
  resolves a gem specified with a pre-release version
  doesn't select a pre-release if not specified in the Gemfile
error: unable to rename temporary '/Users/hsbt/Documents/github.com/bundler/bundler/tmp/tmpdir/_ruby_mjitp36578u581-0e0a36.o-fb2ebbf9' to output file
      '/Users/hsbt/Documents/github.com/bundler/bundler/tmp/tmpdir/_ruby_mjitp36578u581-0e0a36.o': 'No such file or directory'
1 error generated.
  doesn't select a pre-release for sub-dependencies
  selects a pre-release for sub-dependencies if it's the only option
  raises an exception if a child dependency is not resolved
  raises an exception with the minimal set of conflicting dependencies
  should throw error in case of circular dependencies
  should install the latest possible version of a direct requirement with no constraints given
error: unable to rename temporary '/Users/hsbt/Documents/github.com/bundler/bundler/tmp/tmpdir/_ruby_mjitp36578u21-cdf150.o-5817670c' to output file
      '/Users/hsbt/Documents/github.com/bundler/bundler/tmp/tmpdir/_ruby_mjitp36578u21-cdf150.o': 'No such file or directory'
1 error generated.
  should install the latest possible version of a direct requirement with constraints given
  takes into account required_ruby_version
  conservative
    resolves all gems to latest patch
    resolves all gems to latest patch strict
    resolves foo only to latest patch - same dependency case
    resolves foo only to latest patch - changing dependency not declared case
    resolves foo only to latest patch - changing dependency declared case
    resolves foo only to latest patch strict
    resolves bar only to latest patch
    resolves all gems to latest minor
    resolves all gems to latest minor strict
    resolves all gems to latest major
    resolves all gems to latest major strict
    revert to previous
      could revert to a previous version level patch
      cannot revert to a previous version in strict mode level patch
      could revert to a previous version level minor
error: unable to rename temporary '/Users/hsbt/Documents/github.com/bundler/bundler/tmp/tmpdir/_ruby_mjitp36578u29-1ac956.o-5abdcabb' to output file
      '/Users/hsbt/Documents/github.com/bundler/bundler/tmp/tmpdir/_ruby_mjitp36578u29-1ac956.o': 'No such file or directory'
      cannot revert to a previous version in strict mode level minor
1 error generated.


Retried examples: 0

Finished in 1.97 seconds (files took 0.74158 seconds to load)
33 examples, 0 failures

~/D/g/b/bundler (master) > 

The cause is that bundler's test is modifying $TMPDIR and gcc (or probably clang in the above example) is affected by it.

We should call execvpe with TMPDIR=/tmp (and PATH=xxx for cc1, etc.) instead of execvp to start compiler process. I'm not sure about what other environment variables are needed when we spawn the compiler process.

[k0kubun (Takashi Kokubun)]

• Status changed from Open to Closed

It seems that it's no longer reproductive. I guess it's fixed by r62278.

[hsbt (Hiroshi SHIBATA)]

• Assignee set to k0kubun (Takashi Kokubun)
• Backport changed from 2.3: UNKNOWN, 2.4: UNKNOWN, 2.5: UNKNOWN to 2.3: DONTNEED, 2.4: DONTNEED, 2.5: DONTNEED

I also confirmed to fix my error report.