Eric Blake wrote:
> On 04/05/2010 03:08 PM, Bruno Haible wrote:
>> Jim Meyering wrote:
>>> This could be an argument for wrapping some of the C-only tests in a
>>> simple init.sh-using driver (maybe even automatically). Any test that
>>> creates a temporary file would benefit.
>>
>> Yes, I agree: It would make things simpler and more robust if all tests
>> that require temporary files would have a shell script as test driver.
>> Among others, this would guarantee that temporary files get erased when
>> the test gets interrupted with Ctrl-C. In theory, this can be done with
>> a signal handler, but most tests are written in a minimal way and don't
>> care about this. (Only tests/test-stat-time.c and tests/test-posix_spawn3.c
>> install a signal handler for cleanup.)
>
> Many of my recent *at tests use a common idiom of removing such
> temporary files if they happen to exist as garbage from a previous run,
> but I agree that changing things over to a shell wrapper would relieve
> the burden on the C code needing to be robust against Ctrl-C. On the
> other hand, when debugging a test that hangs due to a platform bug, it
> would be nice to keep the temporary files around. So if we do switch to
> shell wrappers, it would be wise to make the cleanup traps use things
> like ${RM=rm} rather than plain rm, to allow the use of RM=: as a way to
> override the cleanup.
init.sh's trap calls its remove_tmp_ function.
If you're debugging a test, you can override that
definition with one that does nothing (say in init.cfg).
