Hi Paul, On 2023-09-03 04:09, Paul Smith wrote: > On Sun, 2023-09-03 at 01:03 +0200, Alejandro Colomar wrote: >> I was also wondering... is ONESHELL significant for performance? >> Does the reduction in number of shells speed up things? > > It might increase performance but that's not what it's for primarily,
For those wondering, here's an experiment with a rule that runs a few
shells (given appropriate values for certain variables):
$ time make -j install DESTDIR=/tmp/foo Z=.gz LINK_PAGES=symlink >/dev/null
real 0m6.229s
user 0m17.820s
sys 0m5.269s
$ sed -i '/ONESHELL/s/^/#/' GNUmakefile
$ time make -j install DESTDIR=/tmp/bar Z=.gz LINK_PAGES=symlink >/dev/null
real 0m8.511s
user 0m23.823s
sys 0m7.511s
It's not orders of magnitude, but it might be interesting to cut a
25%, considering no obvious pitfalls (when you overcome the SHELL
errors, that is :D). For rules with just 1 canned recipe, the times
didn't change.
Here's the rule I used (I chose it because it's the most complex rule
in the Makefile, as it has 3 canned recipes):
$(_manpages):
$(info INSTALL $@)
<$< \
$(SED) $(foreach s, $(MANSECTIONS), \
-e '/^\.so /s, man$(s)/\(.*\)\.$(s)$$, $(notdir
$(man$(s)dir))/\1$(man$(s)ext)$(Z),') \
| $(INSTALL_DATA) -T /dev/stdin $@
ifeq ($(LINK_PAGES),symlink)
if $(GREP) '^\.so ' <$@ >/dev/null; then \
$(GREP) '^\.so ' <$@ \
| $(SED) 's,^\.so \(.*\),../\1,' \
| $(XARGS) -I tgt $(LN) -fsT tgt $@; \
fi
endif
ifeq ($(Z),.bz2)
if ! $(TEST) -L $@; then \
$(BZIP2) $(BZIP2FLAGS) <$@ \
| $(SPONGE) $@; \
fi
else ifeq ($(Z),.gz)
if ! $(TEST) -L $@; then \
$(GZIP) $(GZIPFLAGS) <$@ \
| $(SPONGE) $@; \
fi
else ifeq ($(Z),.lz)
if ! $(TEST) -L $@; then \
$(LZIP) $(LZIPFLAGS) <$@ \
| $(SPONGE) $@; \
fi
else ifeq ($(Z),.xz)
if ! $(TEST) -L $@; then \
$(XZ) $(XZFLAGS) <$@ \
| $(SPONGE) $@; \
fi
endif
> at least not in my opinion. It's there for people who want to use
> shells other than POSIX sh as their SHELL value,
Well, my SHELL is not the POSIX sh, as I'm using bash(1), but I get
it's still POSIX-sh-like, not python. :)
> where it's not so easy
> to create an entire script on a single line and using backslashes.
Can we really call it script if it's invoked via sh -c? It's more
like a command line. :)
Jokes apart, can you put an example of such a scenario? I'm not
imagining it.
BTW, apart from fixing the SHELL (which was a good thing, anyway),
the only change I needed was to remove the '+' from a couple of
recipes (and keep only the first one). Not a big deal. Most rules
work without modification.
Cheers,
Alex
--
<http://www.alejandro-colomar.es/>
GPG key fingerprint: A9348594CE31283A826FBDD8D57633D441E25BB5
OpenPGP_signature
Description: OpenPGP digital signature
