Re: [PATCH 1/6] Makefile: symlink the same way under "symlinks" and "no hardlinks"

[Junio C Hamano <gitster@pobox.com>]

Ævar Arnfjörð Bjarmason  <avarab@gmail.com> writes:

> In particular INSTALL_SYMLINKS=Y will result in a link tree like:
>
>     bin/git
>     libexec/git -> ../bin/git
>     libexec/git-add -> ../bin/git
>
> Whereas NO_INSTALL_HARDLINKS=Y in cases where the "ln" would fail would result in:
>
>     bin/git
>     libexec/git
>     libexec/git-add -> git
>
> I.e. we duplicated the "git" between the bin/ and libexec/
> directories (by default they're hardlinked), and potentially had had
> e.g. "git-add" pointing at the libexec/git hardlink (or more likely if
> "ln" is failing, a copy), instead of the bin/git.
>
> Now we'll instead use the same symlinks to point to the bindir. I
> don't see any reason not to do so, and no breakage related to this has
> been reported with INSTALL_SYMLINKS in all this time. I just did it
> this way to maintain bug-for-bug compatibility at the time.

Makes sense.  I do not see a reason why libexec/git-add that points
at ../bin/git would cause problems, either.

> +# Define NO_INSTALL_HARDLINKS if you'd like to have programs in bin/
> +# and libexec/ either symlinked (we try with INSTALL_SYMLINKS first),
> +# or if that fails fall back on a "cp" instead of a "ln". Useful for
> +# when you don't want hardlinks at all.

So without INSTALL_SYMLINKS and with NO_INSTALL_HARDLINKS, the only
remaining choice is "cp".  With both set, we favour "ln -s" over
"cp" and do not allow "ln".  With INSTALL_SYMLINKS and without
NO_INSTALL_HARDLINKS, we try "ln -s", "ln" and finally "cp".  With
neither, we try "ln" and fall back to "cp"?

That precedence order does make sense.

> @@ -3019,33 +3020,30 @@ endif
>  	} && \
>  	for p in $(filter $(install_bindir_programs),$(BUILT_INS)); do \
>  		$(RM) "$$bindir/$$p" && \
> -		test -n "$(INSTALL_SYMLINKS)" && \
> +		test -n "$(INSTALL_SYMLINKS)" -o "$(NO_INSTALL_HARDLINKS)" && \

I had an impression that we avoid -o/-a used with "test" (instead
we'd use "test && test" or "test || test")?

In either case, if we are told to favor "ln -s", or told not to use
"ln", we try "ln -s"?  That does not make much sense to me, though.

What do I need to do if I do not ever want symbolic links?  Is it
impossible now?

>  		ln -s "git$X" "$$bindir/$$p" || \
>  		{ test -z "$(NO_INSTALL_HARDLINKS)" && \
>  		  ln "$$bindir/git$X" "$$bindir/$$p" 2>/dev/null || \
> -		  ln -s "git$X" "$$bindir/$$p" 2>/dev/null || \
>  		  cp "$$bindir/git$X" "$$bindir/$$p" || exit; }; \
>  	done && \
>  	for p in $(BUILT_INS); do \
>  		$(RM) "$$execdir/$$p" && \
>  		if test -z "$(SKIP_DASHED_BUILT_INS)"; \
>  		then \
> -			test -n "$(INSTALL_SYMLINKS)" && \
> +			test -n "$(INSTALL_SYMLINKS)" -o "$(NO_INSTALL_HARDLINKS)" && \

Perhaps the same comment applies here and to the next hunk, too?

>  			ln -s "$$destdir_from_execdir_SQ/$(bindir_relative_SQ)/git$X" "$$execdir/$$p" || \
>  			{ test -z "$(NO_INSTALL_HARDLINKS)" && \
>  			  ln "$$execdir/git$X" "$$execdir/$$p" 2>/dev/null || \
> -			  ln -s "git$X" "$$execdir/$$p" 2>/dev/null || \
>  			  cp "$$execdir/git$X" "$$execdir/$$p" || exit; }; \
>  		fi \
>  	done && \
>  	remote_curl_aliases="$(REMOTE_CURL_ALIASES)" && \
>  	for p in $$remote_curl_aliases; do \
>  		$(RM) "$$execdir/$$p" && \
> -		test -n "$(INSTALL_SYMLINKS)" && \
> +		test -n "$(INSTALL_SYMLINKS)" -o "$(NO_INSTALL_HARDLINKS)" && \
>  		ln -s "git-remote-http$X" "$$execdir/$$p" || \
>  		{ test -z "$(NO_INSTALL_HARDLINKS)" && \
>  		  ln "$$execdir/git-remote-http$X" "$$execdir/$$p" 2>/dev/null || \
> -		  ln -s "git-remote-http$X" "$$execdir/$$p" 2>/dev/null || \
>  		  cp "$$execdir/git-remote-http$X" "$$execdir/$$p" || exit; } \
>  	done && \
>  	./check_bindir "z$$bindir" "z$$execdir" "$$bindir/git-add$X"