Arch Linux

Why do you want a "|| true" for "hg update"?
> Returns 0 on success, 1 if there are unresolved files.
(man hg, description of update)

Isn't there any way to trap exit code 1 only? "hg pull" returns 255 if something went really wrong, so doing "|| true" to that would be wrong, I think.

@jakobm:
Yes you're right, I went too far.

@engored:
Maybe capturing "hg pull"'s stderr might tell if something really went wrong?

An improved patch is attached.

I was actually thinking about something like this (works for me).

Both patches do not solve the problem for me.

@engored:
Yes, it seems cleaner.

@stefanhusmann:
You also have to take care of the problem in the PKGBUILD itself...

I see. Now makepkg_mercurial2.diff works for me, will test the other one later.

Also the second patch sems to work.

Shouldn't the patch be against pacman/src/pacman-4.0.1/scripts/makepkg.sh.in?

Anyway, I can also confirm this works when combined with modification of the PKGBUILD.

Thanks guys!

So... this is the upstream change:

- Returns 0 on success, 1 if an update had unresolved files.
+ Returns 0 on success, 1 if no changes found or an update had
+ unresolved files.

It seems there is no straightforward way to distinguish genuine failure for lack of changes.

I have a similar problem https://aur.archlinux.org/packages.php?ID=42421 i think caused from mercurial.
An user said that his problem was resolved simply downgrading mercurial to version 2.0.2, so i wonder if at this moment we are looking in the wrong direction for fix this issue.
Because at that point would be a mistake due to the mercurial package and nothing else.
I have this issue on the 64 bit system and don't on the 32 bit system where i tried few day before, but i haven't checked, at that time, which version of mercurial was present on the 32 bit system.
For the moment i haven't tried the solution proposal from that user because i never downgrade a package so i should understand how proceed ( https://wiki.archlinux.org/index.php/Downgrade_packages ) and since at that moment I have no compelling need to use that package if possible i prefer a clean fix.
On summary i hope that the problem resides in the package mercurial.

@Allan:
Yes I think this commit should be reverted.
Would it be conceivable to provide a patched mercurial package?
Alas, some AUR maintainers have already started to change their PKGBUILD (they removed the underscore in the hg variables and append the 'test $? ...' to the 'hg pull' lines) and the irony is: it breaks the build process with mercurial 2.0.2!

Update.
I fixed my problem with the downgrade of mercurial suggested from one user.
This leaves me lean even more than before that this is a bug in the program and that the above solutions proposed (a modification of the makepkg and PKGBUILD on program that use mercurial) are a workaround to that problem but the solution lies in fixing bugs in the program mercurial.

No, the real fix is in makepkg. We really shouldn't care about program exit statuses unless we explicitly catch them. I've already started a branch that does this, but it's 4.1 material, assuming I can convince Dan and Allan that this is a Good Idea™.

Agreed.
If you go in that direction, maybe you could remove the automatic actions performed when _foorepo and _fooroot are defined: I find counterintuitive the fact that a PKGBUILD with an empty 'build' function doesn't do nothing?

The problem here is that we can no longer catch the situation where there is a failure updating an hg repo as mercurial returns the same when the update succeeds but with no file changes. So even if we catch the error explicitly, it ma not be an error...

@Allan
If hg pull is called without -u it will only return 1 if there are no file changes, if hg pull fails it returns 255. Only hg pull -u also returns 1 if the update fails.

So... if I understand hg correctly (big assumption!), this patch should do what we want.

Basically, run "hg pull" and check the return status:
0 - run "hg update"
1 - do nothing
all others - error exit.

Since this requires all mercurial package build scripts be changed, should a separate task be opened against abs to change the PKGBUILD-hg.proto file?

Surprize!
http://selenic.com/hg/rev/a3dcc59054ca
so 'pull' will again return 0.