Re: [gentoo-portage-dev] [0/7] portage_exec cleanups (WAS: [Bug 104705] emerge doesn't print complete error message)

[Jason Stubbs <jstubbs@gentoo.org>]

[-- Attachment #1: Type: text/plain, Size: 2454 bytes --]

On Saturday 29 October 2005 19:34, Jason Stubbs wrote:
> On Sunday 23 October 2005 15:45, Jason Stubbs wrote:
> > Commented on the bug due to reasoning behind this patch. Essentially,
> > SIGTERM is sent to "tee", a WNOHANG waitpid() is performed followed by
> > SIGKILL if it hasn't exited. So if tee doesn't exit immediately upon
> > getting the SIGTERM, its buffers won't get a chance to get to disk due to
> > it being killed. This patch simply adds a 1 second window between the
> > SIGTERM and the SIGKILL.
>
> Per discussion on the bug, the problem is actually that tee shouldn't be
> sent a SIGTERM or SIGKILL at all. The decided solution was to wait on tee
> rather ebuild. I went a bit further than that though and cleaned up most of
> spawn().

Ok, I'm done. The diff is twice the size of the file itself, so I've just 
attached the file. Beyond the previous patch, it's mostly comment additions 
and whitespace/style changes but there are a few edits beyond that:

* Removed unused imports
* Removed the setting of BASH_ENV from spawn_bash
* Removed the SANDBOX_PID_FILE code from spawn_sandbox
* Removed the unnecessary setting of uid in spawn_sandbox
* Added back exception handling to cleanup() as registered pids may
  be cleaned up outside of spawn()
* Changed cleanup() to send a SIGTERM rather than SIGKILL
* Added os.path.isfile() checks to the can-execute tests so that directories
  aren't incorrectly detected as executable
* Changed pathname lookup failure return from None to -1 because
  equatable-to-False returns from spawn() usually indicate success
* Raise a ValueError if a logfile is specified but stdout/stderr have not
* Moved the os.fork()==0 code into a separate function so that it can be
  easily wrapped with a try/except
* Changed the exception reporting to use sys.stderr.write() rather than print
* Moved opt_name and myargs setup into the os.fork()==0 branch as it's not
  used outside of it
* Add/remove spawned pids to/from spawned_pids regardless of whether a
  returnpid was specified
* Changed spawn() to send a SIGTERM rather than SIGKILL when the pipe fails.
* Renamed myc to binary
* Renamed mypid to mypids

Out of those changes, the SIGKILL -> SIGTERM will probably be questioned. 
SIGKILL just seems too heavy handed - especially in the cleanup() function. 
The main thing that worries me is that anything that has write buffers open 
will likely corrupt whatever it was writing to.

--
Jason Stubbs

[-- Attachment #2: portage_exec.py --]
[-- Type: application/x-python, Size: 6078 bytes --]