Re: Strange intermittent LuaJIT errors (likely my fault)

[Alexander Gladysh <agladysh@...>]

On Wed, Apr 11, 2012 at 8:30 PM, Alexander Gladysh <agladysh@gmail.com> wrote:

>> Sigh, not sure what to do. It would definitely help, if the bug
>> could be reproduced determinstically on a single run. Or if the
>> code for the test case could be reduced significantly.

An update:

We're still working on reducing the code base. Not much progress so
far, as removing unused code seems to affect reproducibility of the
bug...

Meanwhile, it looks like the bug is deterministically reproduced on a
fixed n-th run of the suite (this is with somewhat modified code).

For example, on one of the machines, bug is triggered exactly on 111th run.

It seems that some state is kept somewhere between luajit2 runs. Any
ideas *where*?

Each run looks as follows:

strace -o strace.log -e trace=open ./luajit ../../etc/replay.lua data
1 "$@" 2>error.log

That is, a separate executable invocation from the shell script.

https://github.com/Keen-github/luatexts/blob/luajit2-crash-mimimized-err1/test/crash/reproduce.sh#L14

(Also in the news: the bug does not appear on any of our x86_64 machines.)

Alexander.