Writing on Guile optimization: what are anon #x... functions?

Writing on Guile optimization: what are anon #x... functions?

[Artyom Bologov]

Hi y'all,

I'm making a blog post on Guile optimization gotchas I learned. But I
find that one piece of the picture is missing there: anon functions in
the profiler output. Like "anon #x117f408"

One of these functions eats up a lot of performance in my code. So I
can't say I'm an expert on optimization until I understand what it
is. Anyone can hint me at where these functions might come from? Are
they foreign? Are they Guile-internal? Are they not functions at all?

Here's a link to the post draft, where you can find more context (and
maybe correct me if there's something off/missing in the text):

https://www.aartaka.me.eu.org/guile-optimization-gotchas

anon #x117f408 truly is a mystery that haunts me at night...
--
Artyom Bologov

Re: Writing on Guile optimization: what are anon #x... functions?

[Olivier Dion]

On Sat, 06 Apr 2024, Artyom Bologov <mail@aartaka.me> wrote:
> Hi y'all,
>
> I'm making a blog post on Guile optimization gotchas I learned. But I
> find that one piece of the picture is missing there: anon functions in
> the profiler output. Like "anon #x117f408"

Anonymous, i.e. lambda?

[...]
-- 
Olivier Dion
oldiob.ca

Re: Writing on Guile optimization: what are anon #x... functions?

[Artyom Bologov]

Hi Olivier

>Anonymous, i.e. lambda?

No, lambdas display as <current input>:xxx:yyy: where x is lines y is columns. So anon functions are something else.

--
Artyom.

Re: Writing on Guile optimization: what are anon #x... functions?

[Skyler Ferris]

I haven't run into this before in myself, but this line from the NEWS 
file in guile repository seems like a good lead:

"Previously statprof would show strings like "anon #x1234" for primitives
written in C."

It's in the section for changes new to 3.0.3 so if you're on 3.0.2 or 
below then upgrading might help you get more useful analysis. If you're 
already on/past 3.0.3 then I'm confused.

On 4/5/24 13:17, Artyom Bologov wrote:
> Hi y'all,
>
> I'm making a blog post on Guile optimization gotchas I learned. But I
> find that one piece of the picture is missing there: anon functions in
> the profiler output. Like "anon #x117f408"
>
> One of these functions eats up a lot of performance in my code. So I
> can't say I'm an expert on optimization until I understand what it
> is. Anyone can hint me at where these functions might come from? Are
> they foreign? Are they Guile-internal? Are they not functions at all?
>
> Here's a link to the post draft, where you can find more context (and
> maybe correct me if there's something off/missing in the text):
>
> https://www.aartaka.me.eu.org/guile-optimization-gotchas
>
> anon #x117f408 truly is a mystery that haunts me at night...
> --
> Artyom Bologov
>

Re: Writing on Guile optimization: what are anon #x... functions?

[Artyom Bologov]

Hi Skyler,

> "Previously statprof would show strings like "anon #x1234" for primitives
> written in C."

This made me grep through Guile sources. I found this function:

(define (addr->printable addr pdi)
  (or (and=> (and=> pdi program-debug-info-name) symbol->string)
      (and=> (primitive-code-name addr) symbol->string)
      (string-append "anon #x" (number->string addr 16))))

Seems like any function that's neither named in debug info nor a
built-in one is assigned the anon label. So these might be C functions
indeed.

> It's in the section for changes new to 3.0.3 so if you're on 3.0.2 or 
> below then upgrading might help you get more useful analysis. If you're 
> already on/past 3.0.3 then I'm confused.

I'm on Guile 3.0.9, so now I'm confused too!

Thanks,
--
Artyom.