Strong constraints on make install

[Akim Demaille-4]

The GCS require that "make install" does not use

        install FILES... DIR

That's in make-stds:

> @noindent
> Always use a file name, not a directory name, as the second argument  
> of
> the installation commands.  Use a separate command for each file to be
> installed.

I suppose, but I'm not sure, that this is in order to let some
tools postprocess the logs of a "make install".

But that constraints result in bazillions of "install" invocations.
Jim Meyering reported privately that Gettext's "make install"
results in 1200 "install" invocations!!!

That's very expensive, and it can be considerably improved if
the GCS were relaxed on this regard.  There is no portability
issue: install-sh already does the right thing, so the
matter is purely to comply with the standard.

Thanks!

[Karl Berry]

Hi Akim,

    I suppose, but I'm not sure, that this is in order to let some
    tools postprocess the logs of a "make install".

I've never heard of any, but what do I know.  I rather suspect it's
because of different install programs behaved differently, back in the
1980's.

    install-sh already does the right thing,

Right, but do our configure scripts typically check if
install FILE1 FILE2 ... DIR
works, as one criteria for whether install-sh is used?

If so, I can't see any reason to avoid the multiple-file invocation.
If not, it seems like that needs to be done first?

Thanks,
k

[Akim Demaille-4]

Le 20 nov. 07 à 23:40, Karl Berry a écrit :

> Hi Akim,

Hi Karl,

>    I suppose, but I'm not sure, that this is in order to let some
>    tools postprocess the logs of a "make install".
>
> I've never heard of any, but what do I know.

I was suspecting this because of all the magical stuff
around install targets: PRE_INSTALL, NORMAL_INSTALL etc.

> I rather suspect it's
> because of different install programs behaved differently, back in the
> 1980's.

Ah, so it would be purely a matter of portability then.

>    install-sh already does the right thing,
>
> Right, but do our configure scripts typically check if
> install FILE1 FILE2 ... DIR
> works, as one criteria for whether install-sh is used?

No they don't, but of course we'd provide macros for this.
Automake's macros would definitely check that to decide
whether to use the system's install, or intall-sh.

> If so, I can't see any reason to avoid the multiple-file invocation.
> If not, it seems like that needs to be done first?

Sure.  But conversely, in order for Automake to get
updated, included the needed macros, we need the GCS
to accept this change :)

[Karl Berry]

    around install targets: PRE_INSTALL, NORMAL_INSTALL etc.

I don't know if packagers use those as intended.  I personally have
never actually seen them used.
   
    No they don't, but of course we'd provide macros for this.

Ok, fair enough, I'll ask rms.

Thanks,
karl

[Akim Demaille-4]

Le 22 nov. 07 à 00:57, Karl Berry a écrit :

>    around install targets: PRE_INSTALL, NORMAL_INSTALL etc.
>
> I don't know if packagers use those as intended.  I personally have
> never actually seen them used.

Same here.  Maybe this part of the GCS could be relaxed
too?  Such a topic could be discussed on Automake's
list for instance.

[Karl Berry]

    Same here.  Maybe this part of the GCS could be relaxed
    too?  Such a topic could be discussed on Automake's
    list for instance.

Seems like it doesn't hurt to have them, even if they are idle.  Let's
see what rms says about the bigger issue first.

k