Most GnuPG 2.1.1 tests fail on Solaris 10

Dominyk Tiller dominyktiller at gmail.com
Sun May 3 02:55:23 CEST 2015


Hey Werner,

Apologies for the manual copy-paste of an old email discussion. Wanted
to follow up on the latter issue raised.

Homebrew, the OS X package manager, recently switched
(https://github.com/Homebrew/homebrew/pull/39199) its default preferred
TMPDIR over from `/tmp` to the user's own specific tmpdir as defined in
`env`.

We seem to have run into the issue around path length. Or at least, that
*seems* to be the issue. The path is usually around ~50 characters, so
it should be way under the suspected 107, but if I move the `TMPDIR`
default back to `/tmp` and recompile the `make check` stage passes fine
so I'm not quite sure what else it could be.

Full build logs are here:
https://gist.github.com/anonymous/529ffe922261402a75ab

Discussion on the potential breakage and expected end result is here:
https://github.com/Homebrew/homebrew/commit/b95f27ce9f956c56d080b6dc29ba8efb841c6690

Is there anything that can be done to mitigate the breakage other than
either disabling the `make check` stage or reverting the TMPDIR default
change?

Thanks,

Dominyk

Sent from OS X. If you wish to communicate more securely my PGP Public
Key is 0x872524db9d74326c.
=================================================================

On Thu, 18 Dec 2014 13:45, laurent at elanor.org said:

> Well, I hate the backquotes myself, so if you've got a policy to
> replace them with $(), I can manage on my side to replace the shell
> with bash. WONTFIX is fine.

Well, if Solaris is the last of the modern Unices to not support a POSIX
shell, I would indeed tag it as wontfix.

> There's a secret message in there, a gpg-agent error that was not
> displayed, it's really being picky on the path length:

Alright, this is not Solaris specific.  You have this problem on all
Unices because the path length of a Unix domain socket is limit (107
seems to be a common value).

We hafe this socket redirection feature but the API for this also uses
the sun_path field of the sockaddr_un structure.  Thus we have the same
limitations.  But that would also require manual configuration.

Is it really worth to do that or should we better document that
limitation?


Salam-Shalom,

   Werner

-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 801 bytes
Desc: OpenPGP digital signature
URL: </pipermail/attachments/20150503/c3ce8a19/attachment.sig>


More information about the Gnupg-devel mailing list