Re: [Sbcl-devel] More host's gensyms

[Christophe R. <cs...@ca...>]

Manuel Giraud <ma...@le...> writes:

>> Unless there's anything special about any of them, I think one squashed
>> patch is fine.  Fabulous.  Is this still with a slightly customized
>> build (to decrease debug-info) or is this normal?
>
> Thanks! So here it is. I have a customized script to build and diff but
> I've just tested with make.sh and the results are the same.

Thank you, I've pushed this.

>> I wonder also if this would be a reasonable place to add some continuous
>> integration and a baseline to our continuous integration?  44 files
>> isn't 0 but it's not 300 either, and early signals that things have got
>> worse from some commit might help stop us diverging again...
>
> I also have that in mind when we would reach zero… but why not
> before. It could be tricky because the host/platform combination will
> probably change this number.

Yes, it will: we could try to run this on just one host/platform
combination as a baseline.

> While here, I'd like to ask: it seems that many of the differences left
> have to do with the similarity of dumped objects. In
> "compiler/dump.lisp", the similarity-table seems to be an 'equal hash
> table (when cross compiling). So is it possible that the 'equal-ity
> differs from host to host?

EQUAL shouldn't differ (much) from host to host, but differences in the
host compiler's choices about coalescing can leak into the target
(through things like inline function definitions).  What kind of objects
are you seeing?  Do they look like source code?  I wrote about one
problem here:
<http://christophe.rhodes.io/notes/blog/posts/2014/still_working_on_reproducible_builds/>
(it's the second of the three issues).

Cheers,

Christophe