Hi,

On Tue, 25 Aug 2020, Junio C Hamano wrote:

> SZEDER Gábor <szeder.dev@gmail.com> writes:
>
> > I'm afraid I don't understand this patch or the previous one (or
> > both?).  So this new Makefile knob stops hard-linking the dashed
> > builtins _during 'make install'_, but it doesn't affect how Git is
> > built by the default target.  And our CI jobs only build Git by the
> > default target, but don't run 'make install', so setting
> > SKIP_DASHED_BUILT_INS wouldn't have any affect anyway.
>
> Very very true.  Let's drop 3/3 if it is not testing anything new.
>
> I do understand the concern 2/3 wants to address, and it would be a
> real one to you especially if you come from Windows.  People on the
> platform wouldn't be able to use shell scripts written in 12 years
> ago or written with the promise we made to our users 12 years ago,
> and unlike hardlink-capable platforms it incurs real cost to install
> these individual binaries on disk.

Actually, `SKIP_DASHED_BUILT_INS` does not _only_ have an impact on `make
install`:

	$ rm git-add.exe && make
	    BUILTIN git-add.exe
	    BUILTIN all
	    SUBDIR git-gui
	    SUBDIR gitk-git
	    SUBDIR templates

	$ rm git-add.exe && make SKIP_DASHED_BUILT_INS=1
	    BUILTIN all
	    SUBDIR git-gui
	    SUBDIR gitk-git
	    SUBDIR templates

See how `git-add.exe` is linked in the first, but not in the second run?

So the difference 3/3 has is that those hard-linked executables are not
even generated. Now, _technically_ this should not result in any change
because we run the test suite without `--with-dashes`.

Practically, it _does_ make a difference, though, as `bin-wrappers/git`
_specifically_ sets `GIT_EXEC_PATH` to the top-level directory, i.e.
`git-add.exe` _would_ be found if any core Git command that is still
implemented as a script called `git-add`.

Therefore, 3/3 makes sure that we really, really, really do not use those
dashed invocations ourselves.

Ciao,
Dscho