Eric Bavier <bav...@member.fsf.org> writes:
> Could you use a more descriptive phase name? E.g. "patch-sh-and-pwd"?
> [...]
> Ditto. Several other packages use "chdir" or "chdir-to-source"
Done.
> If you know more specifically why parallel builds fail, could you
> comment on that? Race conditions, or not using $(MAKE), etc?
I'm inexperienced with Makefiles, but I suppose the following hints at a
race condition:
--- START SNIP ---
MAKE=make CLISP="/tmp/nix-build-clisp-2.49.drv-0/clisp-2.49/src/clisp -K boot
-E UTF-8 -Epathname 1:1 -Emisc 1:1 -norc" with_dynamic_modules=no ./clisp-link
add boot base i18n syscalls regexp readline || (rm -rf base ; exit 1)
make[1]: Entering directory
'/tmp/nix-build-clisp-2.49.drv-0/clisp-2.49/src/i18n'
make[1]: Nothing to be done for 'clisp-module'.
make[1]: Leaving directory '/tmp/nix-build-clisp-2.49.drv-0/clisp-2.49/src/i18n'
make[1]: Entering directory
'/tmp/nix-build-clisp-2.49.drv-0/clisp-2.49/src/syscalls'
rm -f libgnu_sc.a
rm -f libgnu_sc.a
mkdir -p gllib
mkdir -p gllib
test -f gllib/Makefile || sh config.status gllib/Makefile depfiles
cd gllib; \
make
srcdir="/tmp/nix-build-clisp-2.49.drv-0/clisp-2.49/modules/syscalls/gllib" \
ACLOCAL_M4="/tmp/nix-build-clisp-2.49.drv-0/clisp-2.49/modules/syscalls/gllib/Makefile.am"
\
CFLAGS="-g -O2 -W -Wswitch -Wcomment -Wpointer-arith -Wimplicit
-Wreturn-type -Wmissing-declarations -Wno-sign-compare -Wno-format-nonliteral
-O -falign-functions=4 -DENABLE_UNICODE -DDYNAMIC_FFI -DDYNAMIC_MODULES -I.
-fPIC -I/tmp/nix-build-clisp-2.49.drv-0/clisp-2.49/src/"; \
ln libgnu.a ../libgnu_sc.a
test -f gllib/Makefile || sh config.status gllib/Makefile depfiles
cd gllib; \
make
srcdir="/tmp/nix-build-clisp-2.49.drv-0/clisp-2.49/modules/syscalls/gllib" \
ACLOCAL_M4="/tmp/nix-build-clisp-2.49.drv-0/clisp-2.49/modules/syscalls/gllib/Makefile.am"
\
CFLAGS="-g -O2 -W -Wswitch -Wcomment -Wpointer-arith -Wimplicit
-Wreturn-type -Wmissing-declarations -Wno-sign-compare -Wno-format-nonliteral
-O -falign-functions=4 -DENABLE_UNICODE -DDYNAMIC_FFI -DDYNAMIC_MODULES -I.
-fPIC -I/tmp/nix-build-clisp-2.49.drv-0/clisp-2.49/src/"; \
ln libgnu.a ../libgnu_sc.a
[ SNIP ]
ln: failed to create hard link '../libgnu_sc.a': File exists
Makefile:51: recipe for target 'gllib' failed
make[1]: *** [gllib] Error 1
make[1]: Leaving directory
'/tmp/nix-build-clisp-2.49.drv-0/clisp-2.49/src/syscalls'
Makefile:2282: recipe for target 'base' failed
make: *** [base] Error 1
--- END SNIP ---
This looks like several Makefiles want to ensure the existence of
libgnu_sc.a, and they all start doing it together when called in
parallel.
> Do not begin with "A"; `guix lint` will complain about that. Perhaps
> just "Common Lisp implementation".
Actually 'guix lint' recommended "A Common Lisp implementation", which
is also what GCL in the same file uses. :-)
Maybe lint needs to be fixed then. I went with "Common Lisp
implementation" as you suggested, assuming it's fine that it's more or
less the same with GCL.
> I think in this case the gpl3+ might "win out", but I'll let others
> comment on that.
Done, taking Mark's comment into account.
I pushed the patch already to save a round-trip, assuming that my
decisions above are correct re. race condition comment and synopsis.
Tell me if I did wrong.
Thanks for the review,
Taylan
