Re: [pacman-dev] [arch-dev-public] [signoff] dash 0.5.6-1

On 28/05/10 21:09, Nezmer wrote:

On Fri, May 28, 2010 at 11:17:49AM +0200, Xavier Chantry wrote:

...

On Fri, May 28, 2010 at 10:57 AM, Pierre Schmitz wrote:

...

I would say make sure that core packages are fine with /bin/sh being dash and then move it to testing. This should speed up things. For the install scripts: why can't pacman just call /bin/bash? I mean makpekg and the PKGBUILDS need bash anyway.

I think that the two pending patches from Jonathan Conder on my branch http://code.toofishes.net/cgit/xavier/pacman.git/log/?h=working would make this cleaner and easier to do. 1) switch from popen to execl : popen implicitely calls /bin/sh, so when we had /bin/sh=dash and wants bash, it would call /bin/dash then /bin/bash. But execl gives us more control. 2) ldconfig is now called directly so if we do the change /bin/sh -> /bin/bash, it only affects scriptlet executions.

But I am pretty sure we talked about this before, maybe there was a bug report, and some people suggested to make the shell path configurable. Thinking about it now, I would be fine with just calling /bin/bash directly. If we consider pacman project as a whole, it's pretty much tied to bash anyway.

+1 from someone who is using pacman in a FreeBSD environment. Where "/bin/sh" is linked to neither bash nor dash.

bash is used in both makepkg and repo-add. It's not a platform-dependent terrible dependency. And for recovery purposes, It can optionally be built statically. The static option is actually supported as a configure flag.

People who use pacman do not necessarily use makepkg or repo-add so using that as a basis for hard coding /bin/bash is wrong. Dan and I have discussed allowing a #! at the top of install scripts. I think he had some very draft code that even allowed python scriptlets! Allan