On January 7, 2018 4:18 PM, brian m. Carlson wrote: > On Sun, Jan 07, 2018 at 03:57:59PM -0500, Randall S. Becker wrote: > > I'm looking for a proper (i.e. not sneaky) way to detect the platform > > I am on during testing so that some tests can be modified/skipped > > other than using the standard set of dependencies. In particular, the > > maximum path on current NonStop platforms is 8-bit 2048 bytes. It > > appears that there are some tests - at least from my preliminary > > "guessing" - that are beyond that limit once all of the path segments > > are put together. I would rather have something in git that specifies > > a path size limit so nothing exceeds it, but that may be wishing. > > The way we usually skip tests automatically is with a test prerequisite. > You might look at t/test-lib.sh for the test_set_prereq and test_lazy_prereq > calls and synthesize one (maybe LONG_PATHS) that meets your needs. You > can then annotate those tests with the appropriate prerequisite. > > I expect that for long paths, you will hit a lot of the same issues as occur on > Windows, where PATH_MAX may be very small. It might be valuable to > expose this information as a build option and then set an appropriate > variable in t/test-lib.sh. Where I am, at this point: I have PATH_MAX defined in Makefile as optional and which can be specified as a number in config.mak.uname. If provided, it adds -DPATH_MAX to BASIC_CFLAGS, which will ensure consistency with limits.h (if the values are different, at least c99 warns about it). I've also got it into GIT-BUILD-OPTIONS, if defined. From there it seems straight-forward to use it in test scripts using standard shell scripting, however, I can't find a good model/function for what would be a prerequisite check consistent with existing git test methods - you know, clarity. One approach I have been pursuing is to use test_set_prereq if PATH_MAX is defined, and add a new method like test_missing_prereq_eval that would take PATH_MAX and an expression, like -le 2048, to cause a test to be skipped if the variable is defined but the evaluation fails. I'm still having noodling through trying to make that work, and if anyone has a better idea (please have a better idea!!), please please suggest it. Cheers, Randall -- Brief whoami: NonStop&UNIX developer since approximately UNIX(421664400)/NonStop(211288444200000000) -- In my real life, I talk too much.
