Skip to Content.
Sympa Menu

sm-discuss - Re: [SM-Discuss] Watch out for installwatch change: commit f9349ce4743d31082f18b193174d35089fce69b3

sm-discuss AT lists.ibiblio.org

Subject: Public SourceMage Discussion List

List archive

Chronological Thread  
  • From: Eric Sandall <eric AT sandall.us>
  • To: sm-discuss AT lists.ibiblio.org
  • Subject: Re: [SM-Discuss] Watch out for installwatch change: commit f9349ce4743d31082f18b193174d35089fce69b3
  • Date: Wed, 9 Sep 2020 08:49:55 -0700

On 8/31/20 09:03, Thomas Orgis wrote:
Aloha,

I am still not giving up, perhaps only to entertain ar on IRC. I just
pushed a change that might be controversial and wrong, so I announce
it here. Commit f9349ce4743d31082f18b193174d35089fce69b3 adds a patch
to installwatch that drops a set of read-only functions from it.

The reason is twofold:

1. My installation is seriously damaged and misses some sorcery state
that causes castfs installs to stumble and fail upon existing files of
the spell which is not recorded as installed and hence not deinstalled
before writing new stuff.*

2. Switching to installwatch instead of castfs works as such, but gets
stuck in a deadlock while installing rust. Rust contains jemalloc,
which calls readlink() during intialization. That prompted
installwatch, which then tried to allocate some memory in a libc call …
which then went into jemalloc, which is waiting for its initialization
… nasty.

So installwatch is wrapping less now. Anyone got an issue with that?
The list of installed files still seems to be captured just fine.

Maybe we want to make a new tarball for installwatch instead of
collecting patches. Even if there is an overlayfs replacement for
castfs, I think we need a lightweight fallback like installwatch, or
maybe even just something based on strace -ff.


Alrighty then,

Thomas


* Needless to say, I'd consider it much saner to _not_ delete the old
  files before writing the new ones, for a smoother transition where
  there is minimal time window of libraries missing. Of course, there
  is still different breakage possible with inconsistent states. We'd
  need atomic changesets in the root filesystem.


Thank you for working on this Thomas. :) I've had to set STAGED_INSTALL="off" to some spells to work around our older castfs, which usually works great, but not always.

-sandalle

Attachment: signature.asc
Description: OpenPGP digital signature



  • Re: [SM-Discuss] Watch out for installwatch change: commit f9349ce4743d31082f18b193174d35089fce69b3, Eric Sandall, 09/09/2020

Archive powered by MHonArc 2.6.24.

Top of Page