On Sat, Feb 17, 2018, 6:38 AM Stefan Sperling <s...@apache.org> wrote:
> Hi Troy, > > The diff --normal option is a GNU extension and hence not portable to > systems which do not use GNU diff. > > See > https://ci.apache.org/builders/svn-bb-openbsd/builds/27/steps/Test%20bindings/logs/stdio > which fails with: > > trunk/subversion/bindings/swig/python/tests/run_all.py > ............................................diff: unknown option -- normal > > I would suggest to just drop the --normal option if possible, because it > seems to be the default behaviour for GNU diff and likely also other > implementations. > Ah, well that is what I get for just looking at the help output on my box! Honestly it is really cruft from my first cut at the test, where it took having a non-empty list to actually call the executable (versus the current None check). > Another suggestion: Is there a pythonic way to test for the existence > of a 'diff' program, instead of assuming that any system except Windows > has one installed? > Our main test suite has a concept of 'skipped' tests, and if this exists > in the python bindings tests as well it would make sense to report the > test result as 'skipped' instead of 'pass' if no diff tool is available. > I did look at this for a bit since I hated the "not win32" check, especially since you *could* have diff on windows. However, it appeared there was quite a bit of code to reimplement, and all the examples I found still had something wrong with them for particular scenarios. But that search was pretty windows-centric, and not particularly exhaustive. It seems it is worth another look. Surely there is a way to just try calling it, and catch an exception or something similar without reimplementing searching through the path and using platform-specific calling conventions (like testing with and without .exe). I'm research a better method for this and come up a solution that is a bit more general. Troy
