Re: profiler-report seems to be missing data?

[Stefan Monnier <monnier@iro.umontreal.ca>]

> Byte compiling the file didn't seem to have much effect: let* was still
> blamed for the vast majority of the runtime (57%).

`let*` does not exist in byte-compiled code (it's decomposed into
various other byte-codes instead), so its presence in the profile
indicates that somehow non-compiled code is executed.

>    - The overall number of CPU samples didn't change, I think indicating
>    that eliminating the let* didn't in fact speed up the code. (In both cases,
>    I ran ftf-find-file once, which took 4-5 seconds.)
>    - The samples that were originally blamed on let* are instead blamed
>    largely on cons and gethash now, which in my opinion seems more likely
>
> My suspicion here is that let* and the profiler are having some bad
> interaction where the samples are incorrectly being attributed to let* when
> they should be instead attributed to code called within the let*. Stefan,
> do you think I'm interpreting this correctly?

Sounds possible, yes.  Not sure if it's specific to `let*`, tho.
[ E.g. maybe it's just that the leaf of each stack trace gets lost?  ]


        Stefan