[PATCH] 18969 - multiple string test failures due to missing locale dependencies

[Florian Weimer]

On 10/12/2015 06:43 PM, Mike Frysinger wrote:

> so it creates one dep per locale.  the rule for processing these targets:
> $(addprefix $(common-objpfx)localedata/,$(CTYPE_FILES)): %: \
>   ../localedata/gen-locale.sh \
>   $(common-objpfx)locale/localedef \
>   ../localedata/Makefile \
>   $(addprefix ../localedata/charmaps/,$(CHARMAPS)) \
>   $(addprefix ../localedata/locales/,$(LOCALE_SRCS))
>     @$(SHELL) ../localedata/gen-locale.sh $(common-objpfx) \
>           '$(built-program-cmd-before-env)' '$(run-program-env)' \
>           '$(built-program-cmd-after-env)' $@; \
>     $(evaluate-test)
> 
> so it generates one locale per invocation.  it looks correct to me.
> 
> then again, it's fairly easy to test: just go into one of these dirs,
> delete the locale output, and run `make -j`.  see if any of the commands
> are repeated.

I still see occasional crashes in the localedata tests.  Your “make -j”
check does not work because the action in the rule above is prefixed
with “@”.  Can we remove that, so that we may end up with more evidence
of unsynchronized, parallel execution?

Florian