Hi,
I recently uploaded Boost 1.54, which initially built fine on mipsel (used the
buildd eysler). Five days later, I uploaded again with some small fixes and
the build failed on mipsel. It didn't just fail, but it failed spectacularly
with multiple Internal Compiler Errors -- see #715526.
Later, I read an interesting thread about "DSA Concerns", which included this
tidbit:
* mipsel: the porter machine and some of the buildd machines have an
implementation error for one opcode; missing kernel in the archive [1]
Is this referring to "mayer"? The ensuing thread contained some references to
different machine types such as "2e". Eysler is running kernel "Linux 3.2.0-4-
loongson-2e" while Mayer is running "3.2.0-4-sb1-bcm91250a". Since the buildd
for a given build is selected at random, I elected to re-try the build on a
porterbox, eder. This one is also a "loongson-2e" and the build succeeded.
So the evidence I have is that "2e" machines work and "bcm91250a" machines do
not. I'm open to suggestions as to #715526: is the ICE truly a compiler bug?
Is it a bug in the kernel? Other part of the toolchain? Is the physical
hardware flakey?
Can we remove mayer so that packages don't artificially fail to build?
Thanks,
-Steve
[1] https://lists.debian.org/debian-mips/2013/06/msg00048.html
Attachment:
signature.asc
Description: This is a digitally signed message part.