Re: [PATCH 1/2] make: add install-stripped target
- From
Junio C Hamano <gitster@pobox.com>
- Date
- Aug 26, 2021, 20:08 UTC
- Message-ID
- <xmqq8s0odl5l.fsf@gitster.g>
- In-Reply-To
- <20210826113824.50078-2-bagasdotme@gmail.com>
Bagas Sanjaya <bagasdotme@gmail.com> writes:
Show 5 quoted lines
> +install-stripped: install
> + for f in $(PROGRAMS) git$X; do \
> + find $$prefix -type f -name $$f -exec $(STRIP) $(STRIP_OPTS) {} \; ; \
> + done
> +This sounds awfully wasteful.
The recipe for the install target knows exactly each of these programs are installed, but yet the above is running around inside $prefix to find them after the fact.
It also looks incorrect, too.
It is not guaranteed that $prefix does not contain any $IFS whitespace in it, and worse yet, $prefix may not contain $bindir or $libexecdir in it, so find may never reach these binaries.
It also depends on "strip" not to break handlinks to the same binary. "git" is linked to many built-in command binary like "git-cat-file" and "git-remote-$curl" for various protocols are installed by creating links to "git-remote-http". It seems that the "strip" command from GNU binutils package strips such a binary in-place, but I do not think there is no fundamental reason to believe that everybody else's "strip" would behave that way.
I would have expected that 'install-stripped' and 'install' targets would run the same recipe, and when $(install_bindir_programs) are installed in $(bindir) using $(INSTALL), we would optionally pass the '--strip' option to the $(INSTALL) program when the recipe is run for the install-stripped target. All the tricky symlinking, hardlinking and copying happens only on the result of that step, and the strip step should happen before that, I would think.
> +.PHONY: install-gitweb install-doc install-man install-man-perl install-html install-info install-pdf install-stripped
Split the overly long line like this into two or more.
> .PHONY: quick-install-doc quick-install-man quick-install-html > install-gitweb: > $(MAKE) -C gitweb install