Eric Blake wrote: > I just noticed a problem. update-copyright is a perl script only intended for > maintainer use, which is fine, since only maintainers are required to have > perl > installed (after all, we already know that maintainers have perl by virtue of > the fact that they are running autoconf). But, as explicitly permitted by the > GNU standards, end users need not have good enough perl. In other words, > blindly including test-update-copyright as part of 'make check' in any package > that bootstrapped via 'gnulib-tool --with-tests' can lead to spurious failures > for some end users because the test is assuming more than the end user is > required to have. > > One option is to just relax test-update-copyright.sh to gracefully skip the > test if perl is not found or can't successfully perform copyright updates. > Another option is to figure out a way to make gnulib tool distinguish between > tests of end-user code (most tests) vs. tests of maintainer-only tools (test- > update-copyright, test-vc-list-files, ...) and set things up so that > maintainer > tool tests are only run by 'make distcheck' rather than 'make check'. > Actually, even 'make distcheck' might be wrong, since that is supposed to work > without autotools; and 'make maintainer-check' as part of maint.mk may be a > better place. Any other ideas, or an approach for a patch along any of these > lines? > > For the record, I discovered this because 'perl -ei' fails on cygwin 1.5 [but > works in cygwin 1.7] because of older cygwin's inability to delete in-use > files:
That should be easy to detect. Do you mean that perl -pi fails on cygwin 1.5? For example, would this fail? echo a > f && perl -pi -e s/a/b/ f
