[arch-general] pacman/libalpm/libfetch do not honor TMPDIR

[clemens fischer]

clemens fischer wrote:

> Gaetan Bisson wrote:
> 
>> If that's easy then it shouldn't be too hard for you to open a bug
>> report on the tracker and submit a patch.
> 
> Then what component does the actual extracting?  Is it libfetch?  That
> would be an upstream moving target, because it comes from netbsd.  Do
> you guys accept patches against libfetch in this case?

Seems libarchive is used to handle packages.

pacman.git lib/libalpm/add.c::perform_extraction() is hard to
understand.  It should be the place the warning I cited came from.  If
that is the case, ENOSPC wasn't the cause of it.

I don't understand two things:

- where exactly are files extracted to?  If they go to /, how would
  people use pacman when root is mounted ro (in the case of eg.
  /.INSTALL)?  If I can freely determine where files go via
  archive_entry_set_pathname(), where's the problem in using the value
  of TMPDIR as a prefix and move the file into its final position when
  libalpm declares the transaction ok?

- the file causing the warning was ".INSTALL".  It contains
  upgrade/install scriptlets.  How can a problem extracting an important
  file like this be considered a warning?


clemens