Re: [PATCH 0/6] build-aux/bootstrap that doesn't need to replace itself

[Paul Eggert]

On 12/27/22 13:09, Bruno Haible wrote:
> I strongly object against the backwards move regarding the autopull.sh /
> autogen.sh *concepts*.

The concepts are still there; the only issue is the API. Whether the API 
should be './bootstrap --pull && ./bootstrap --gen' or './autopull.sh && 
./autogen.sh' is a relatively minor detail, as far as the 
pull-vs-generate concepts are concerned.

When the API was proposed I put it on my list of things to worry about, 
and unfortunately I didn't get around to thinking about it seriously 
until recently. Having done so, I took the trouble of implementing and 
suggesting a API that seems to be better for developers of gzip, tar and 
I assume other packages.

Even if we can't agree whether the newer API is better enough to 
standardize on, that should be OK. People doing software imports already 
have to deal with dozens of such APIs, and whether there's one vs two 
more APIs won't make much difference. That being said, if it matters to 
have just one API for Gnulib-using apps, I can volunteer to help migrate 
any packages already using the autogen.sh/autopull.sh API; it shouldn't 
be that hard, and these packages can even continue to support both APIs 
if that is needed for some reason.

>      it encourages packages to clump two different things into a single
>      script.

That shouldn't be much of a problem, any more that it's much of a 
problem that "git fetch" and "git diff" are in a single "git" command.

The current implementation is not cast in stone, and we can to work to 
make it better, more flexible, improve the documentation, etc.

PS. I got one private email from someone else in response to the recent 
patches, in that they didn't like the name "autogen". Not sure what name 
they'd prefer or why, but have asked.