| Subject: | The recent blib/_eumm/ addition fires too early breaking legitimate dev workflows |
When working on pure-perl dists, it is relatively common to spot check with `prove -l <set of tests>` without doing a full `make` run. The creation of blib/_eumm now happens at `perl Makefile.PL` time (instead of the more reasonable `make pm_to_blib`), leaving a stub blib/ with no ...lib/ undernath it. This in turn breaks any workflow which checks of blib/ existing before deciding what to push to @INC or something similar: http://grep.cpan.me/?q=-%5Bed%5D%5Cs%2B%28%27blib%27%7C%22blib%22%29
To put it differently, the following workflow:
git clone ...
cd ...
perl Makefile.PL
prove -lr
No longer works as intended for most of the distributions listed in the above cpangrep (either by silently not processing anything in `lib` because an empty `blib` exists, or by outright death).
Related workaround for an affected CI setup: https://github.com/dbsrgits/dbix-class/commit/e95056151