Re: mkinstalldirs --version

[Akim Demaille]

>>> "Bruno" == Bruno Haible <address@hidden> writes:
 > [...]

 >>> Bruno, do you think setting
 >>> mkinstalldirs = $(install_sh) -d
 >>> would cause any issue?

 Bruno> The only immediate issue that I can see is that invoking
 Bruno> $(mkinstalldirs) would become slower: 1. sh has to parse
 Bruno> 4 times more shell code,

 > :)

 Bruno> 2. it doesn't have the "mkdir -p" optimization that
 Bruno> mkinstalldirs has.

 > Then, how about a configure test to set `mkinstalldirs = mkdir -p'
 > when it works?  (This, of course, is independent of whether the
 > replacement is mkinstalldirs or install-sh.)

In the first place, I, for one, am not convinced by the usefulness of
creating the directories, and then installing in them.  How about this
old plan of extending AC_PROG_INSTALL with checking that install can
create directories on the fly?  We can still keep the create-dir
targets for those willing to be able to create this dirs without
populating them, but the install targets would not need to depend on
this.  Using good "install" program would be even a better speed up
than that of 1/4 of shell code.