Re: Unnecessary memoization invalidation

Mathew,

Please go ahead and do a PR per issue and we can review the proposed
changes there.

-Bill

On Mon, Dec 9, 2019 at 9:30 AM Mathew Robinson <[email protected]>
wrote:

> Hello all,
>
> While doing build system performance work at MongoDB I've found a few
> areas where the memoization isn't shared between threads (similar to the
> ninfo bug I resolved a few months back) and most importantly that
> Node._clear_memoized_values is being called a little too aggressively.
>
> By removing a few call sites of Node._clear_memoized_values I was able to
> remove a few thousand extra calls to stat which had perf benefits for our
> build. I didn't measure the impact on every memoized value. Given the
> sensitivity and difficulty of caching I wanted to reach out here before
> sending a PR.
>
> Additionally I noticed that for commands that don't change and in WhereIs
> there is no memoization. This causes ~60 - 100 thousand unnecessary stat
> calls for our builds. I'd like to memoize WhereIs at a minimum as well,
> though that change is unrelated to the above.
>
> Thanks,
> - Mathew Robinson @chasin glogic
> _______________________________________________
> Scons-users mailing list
> [email protected]
> https://pairlist4.pair.net/mailman/listinfo/scons-users
>

_______________________________________________
Scons-users mailing list
[email protected]
https://pairlist4.pair.net/mailman/listinfo/scons-users