Re: [Caml-list] improve omake [was One build system to rule them all]

[Mark Shinwell <mshinwell@janestreet.com>]

On 23 September 2014 11:40, Alain Frisch <alain@frisch.fr> wrote:
> On 09/19/2014 05:18 PM, Yaron Minsky wrote:
>>
>> - The build itself has limited parallelism because of various
>>    bottlenecks inside omake, like the fact that it computes its md5sums
>>    in a single process.
>
> I think that omake doesn't usually recompute the md5 when the file hasn't
> changed (probably based on the mtime, maybe inode as well), so I'm surprised
> by the claim.  Did you actually confirm that the computation of md5 was one
> of the bottlenecks or is it just an hypothesis?

Without wishing to prolong a discussion about improving omake...

I think the details are hazy now, but as far as we recall: your
statement about avoiding recomputation is correct, but a lot of md5sum
work happened subsequent to each compilation action in serial, within
the omake process.

By no means was this the only problem; another, for example, was a
vast amount of time spent constructing rules for an entire tree even
if only a small portion was being demanded to be built.

(Of course, the worst problem with omake was probably the DSL, which
we found to be wholly unsuitable for large-scale reliable rule
engineering.  Programmability really is key.)

Mark