[cocci] Very long compile times on parsing_cocci/parser_cocci_menhir.ml

[cocci] Very long compile times on parsing_cocci/parser_cocci_menhir.ml

[Richard W.M. Jones]

OCAMLOPT  parsing_cocci/parser_cocci_menhir.ml

This single file takes a very long time to compile, about 34 minutes
on my 2 year old Intel laptop.  It's quite large too ...

139201 ./coccinelle-1.2/parsing_cocci/parser_cocci_menhir.ml

I've observed in the past that the OCaml compiler doesn't handle
single large files well.  It seems to have some O(n^k) behaviour.

Is this a known issue?

Rich.

coccinelle 1.2 from tarball
ocaml-5.1.1-4.fc40.x86_64
ocaml-menhir-20231231-3.fc40.x86_64

-- 
Richard Jones, Virtualization Group, Red Hat http://people.redhat.com/~rjones
Read my programming and virtualization blog: http://rwmj.wordpress.com
virt-top is 'top' for virtual machines.  Tiny program with many
powerful monitoring features, net stats, disk stats, logging, etc.
http://people.redhat.com/~rjones/virt-top

Re: [cocci] Very long compile times on parsing_cocci/parser_cocci_menhir.ml

[Wolfram Sang]

[-- Attachment #1: Type: text/plain, Size: 335 bytes --]

Hi Rich,

> Is this a known issue?

I can't confirm. On my laptop from 2017 (2 cores, 8 GB RAM), compiling
from latest git (after 'make clean') takes 90 seconds.

OCaml: The OCaml toplevel, version 4.14.1
Menhir: menhir, version 20231231

Both from Debian Testing.

Or did I miss a step?

Happy hacking,

   Wolfram


[-- Attachment #2: signature.asc --]
[-- Type: application/pgp-signature, Size: 833 bytes --]

Re: [cocci] Very long compile times on parsing_cocci/parser_cocci_menhir.ml

[Julia Lawall]

On Thu, 2 May 2024, Richard W.M. Jones wrote:

> OCAMLOPT  parsing_cocci/parser_cocci_menhir.ml
>
> This single file takes a very long time to compile, about 34 minutes
> on my 2 year old Intel laptop.  It's quite large too ...
>
> 139201 ./coccinelle-1.2/parsing_cocci/parser_cocci_menhir.ml
>
> I've observed in the past that the OCaml compiler doesn't handle
> single large files well.  It seems to have some O(n^k) behaviour.
>
> Is this a known issue?

I agree that it's slow, but I haven't seen it take 34 minutes.

The point is that it is a parser, so the code is generated.  I use OCaml
4.13.1 or 5.0, not 5.1.1.  I have the same version of menhir as the one
you mention.

I can try to make it smaller.

julia

>
> Rich.
>
> coccinelle 1.2 from tarball
> ocaml-5.1.1-4.fc40.x86_64
> ocaml-menhir-20231231-3.fc40.x86_64
>
> --
> Richard Jones, Virtualization Group, Red Hat http://people.redhat.com/~rjones
> Read my programming and virtualization blog: http://rwmj.wordpress.com
> virt-top is 'top' for virtual machines.  Tiny program with many
> powerful monitoring features, net stats, disk stats, logging, etc.
> http://people.redhat.com/~rjones/virt-top
>
>

Re: [cocci] Checking compilation durations for “parser_cocci_menhir.ml”

[Markus Elfring]

> This single file takes a very long time to compile, about 34 minutes
> on my 2 year old Intel laptop.

I find such an experience report surprising.
How do you think about to improve software profiling accordingly?


> It's quite large too ...
>
> 139201 ./coccinelle-1.2/parsing_cocci/parser_cocci_menhir.ml

How will the software evolve further here?


> I've observed in the past that the OCaml compiler doesn't handle
> single large files well.  It seems to have some O(n^k) behaviour.

Did you notice any differences between compiler versions?

Regards,
Markus

Re: [cocci] Very long compile times on parsing_cocci/parser_cocci_menhir.ml

[Richard W.M. Jones]

On Thu, May 02, 2024 at 11:01:14AM +0200, Wolfram Sang wrote:
> Hi Rich,
> 
> > Is this a known issue?
> 
> I can't confirm. On my laptop from 2017 (2 cores, 8 GB RAM), compiling
> from latest git (after 'make clean') takes 90 seconds.
> 
> OCaml: The OCaml toplevel, version 4.14.1

I think there is some significant difference between OCaml 4.x and
OCaml 5.1.1.  Can you see if updating to OCaml 5.1.1 makes it slower?

Rich.

> Menhir: menhir, version 20231231
> 
> Both from Debian Testing.
> 
> Or did I miss a step?
> 
> Happy hacking,
> 
>    Wolfram
> 



-- 
Richard Jones, Virtualization Group, Red Hat http://people.redhat.com/~rjones
Read my programming and virtualization blog: http://rwmj.wordpress.com
virt-p2v converts physical machines to virtual machines.  Boot with a
live CD or over the network (PXE) and turn machines into KVM guests.
http://libguestfs.org/virt-v2v

Re: [cocci] Very long compile times on parsing_cocci/parser_cocci_menhir.ml

[Richard W.M. Jones]

On Thu, May 02, 2024 at 11:44:00AM +0200, Julia Lawall wrote:
> 
> 
> On Thu, 2 May 2024, Richard W.M. Jones wrote:
> 
> > OCAMLOPT  parsing_cocci/parser_cocci_menhir.ml
> >
> > This single file takes a very long time to compile, about 34 minutes
> > on my 2 year old Intel laptop.  It's quite large too ...
> >
> > 139201 ./coccinelle-1.2/parsing_cocci/parser_cocci_menhir.ml
> >
> > I've observed in the past that the OCaml compiler doesn't handle
> > single large files well.  It seems to have some O(n^k) behaviour.
> >
> > Is this a known issue?
> 
> I agree that it's slow, but I haven't seen it take 34 minutes.

I just tried it on a much faster AMD desktop machine, but essentially
the same OCaml and menhir, and it still takes a very long time to
compile this one file: 17:35 minutes

I tried it on a (different) machine with OCaml 4.11.1 and that was
considerably faster, taking under a minute.

Rich.

> The point is that it is a parser, so the code is generated.  I use OCaml
> 4.13.1 or 5.0, not 5.1.1.  I have the same version of menhir as the one
> you mention.
> 
> I can try to make it smaller.
> 
> julia
> 
> >
> > Rich.
> >
> > coccinelle 1.2 from tarball
> > ocaml-5.1.1-4.fc40.x86_64
> > ocaml-menhir-20231231-3.fc40.x86_64
> >
> > --
> > Richard Jones, Virtualization Group, Red Hat http://people.redhat.com/~rjones
> > Read my programming and virtualization blog: http://rwmj.wordpress.com
> > virt-top is 'top' for virtual machines.  Tiny program with many
> > powerful monitoring features, net stats, disk stats, logging, etc.
> > http://people.redhat.com/~rjones/virt-top
> >
> >

-- 
Richard Jones, Virtualization Group, Red Hat http://people.redhat.com/~rjones
Read my programming and virtualization blog: http://rwmj.wordpress.com
virt-top is 'top' for virtual machines.  Tiny program with many
powerful monitoring features, net stats, disk stats, logging, etc.
http://people.redhat.com/~rjones/virt-top

Re: [cocci] Very long compile times on parsing_cocci/parser_cocci_menhir.ml

[Julia Lawall]

On Thu, 2 May 2024, Richard W.M. Jones wrote:

> On Thu, May 02, 2024 at 11:44:00AM +0200, Julia Lawall wrote:
> >
> >
> > On Thu, 2 May 2024, Richard W.M. Jones wrote:
> >
> > > OCAMLOPT  parsing_cocci/parser_cocci_menhir.ml
> > >
> > > This single file takes a very long time to compile, about 34 minutes
> > > on my 2 year old Intel laptop.  It's quite large too ...
> > >
> > > 139201 ./coccinelle-1.2/parsing_cocci/parser_cocci_menhir.ml
> > >
> > > I've observed in the past that the OCaml compiler doesn't handle
> > > single large files well.  It seems to have some O(n^k) behaviour.
> > >
> > > Is this a known issue?
> >
> > I agree that it's slow, but I haven't seen it take 34 minutes.
>
> I just tried it on a much faster AMD desktop machine, but essentially
> the same OCaml and menhir, and it still takes a very long time to
> compile this one file: 17:35 minutes
>
> I tried it on a (different) machine with OCaml 4.11.1 and that was
> considerably faster, taking under a minute.

I tried it on an Intel server and it takes a few seconds.  OCaml version
5.1.1.  menhir, version 20200123

julia

>
> Rich.
>
> > The point is that it is a parser, so the code is generated.  I use OCaml
> > 4.13.1 or 5.0, not 5.1.1.  I have the same version of menhir as the one
> > you mention.
> >
> > I can try to make it smaller.
> >
> > julia
> >
> > >
> > > Rich.
> > >
> > > coccinelle 1.2 from tarball
> > > ocaml-5.1.1-4.fc40.x86_64
> > > ocaml-menhir-20231231-3.fc40.x86_64
> > >
> > > --
> > > Richard Jones, Virtualization Group, Red Hat http://people.redhat.com/~rjones
> > > Read my programming and virtualization blog: http://rwmj.wordpress.com
> > > virt-top is 'top' for virtual machines.  Tiny program with many
> > > powerful monitoring features, net stats, disk stats, logging, etc.
> > > http://people.redhat.com/~rjones/virt-top
> > >
> > >
>
> --
> Richard Jones, Virtualization Group, Red Hat http://people.redhat.com/~rjones
> Read my programming and virtualization blog: http://rwmj.wordpress.com
> virt-top is 'top' for virtual machines.  Tiny program with many
> powerful monitoring features, net stats, disk stats, logging, etc.
> http://people.redhat.com/~rjones/virt-top
>
>

Re: [cocci] Very long compile times on parsing_cocci/parser_cocci_menhir.ml

[Richard W.M. Jones]

On Thu, May 02, 2024 at 06:39:34PM +0200, Julia Lawall wrote:
> 
> 
> On Thu, 2 May 2024, Richard W.M. Jones wrote:
> 
> > On Thu, May 02, 2024 at 11:44:00AM +0200, Julia Lawall wrote:
> > >
> > >
> > > On Thu, 2 May 2024, Richard W.M. Jones wrote:
> > >
> > > > OCAMLOPT  parsing_cocci/parser_cocci_menhir.ml
> > > >
> > > > This single file takes a very long time to compile, about 34 minutes
> > > > on my 2 year old Intel laptop.  It's quite large too ...
> > > >
> > > > 139201 ./coccinelle-1.2/parsing_cocci/parser_cocci_menhir.ml
> > > >
> > > > I've observed in the past that the OCaml compiler doesn't handle
> > > > single large files well.  It seems to have some O(n^k) behaviour.
> > > >
> > > > Is this a known issue?
> > >
> > > I agree that it's slow, but I haven't seen it take 34 minutes.
> >
> > I just tried it on a much faster AMD desktop machine, but essentially
> > the same OCaml and menhir, and it still takes a very long time to
> > compile this one file: 17:35 minutes
> >
> > I tried it on a (different) machine with OCaml 4.11.1 and that was
> > considerably faster, taking under a minute.
> 
> I tried it on an Intel server and it takes a few seconds.  OCaml version
> 5.1.1.  menhir, version 20200123

Interesting, thanks Julia.  I'll poke this further to see
if I can find out what the causative factor is.

Rich.

> julia
> 
> >
> > Rich.
> >
> > > The point is that it is a parser, so the code is generated.  I use OCaml
> > > 4.13.1 or 5.0, not 5.1.1.  I have the same version of menhir as the one
> > > you mention.
> > >
> > > I can try to make it smaller.
> > >
> > > julia
> > >
> > > >
> > > > Rich.
> > > >
> > > > coccinelle 1.2 from tarball
> > > > ocaml-5.1.1-4.fc40.x86_64
> > > > ocaml-menhir-20231231-3.fc40.x86_64
> > > >
> > > > --
> > > > Richard Jones, Virtualization Group, Red Hat http://people.redhat.com/~rjones
> > > > Read my programming and virtualization blog: http://rwmj.wordpress.com
> > > > virt-top is 'top' for virtual machines.  Tiny program with many
> > > > powerful monitoring features, net stats, disk stats, logging, etc.
> > > > http://people.redhat.com/~rjones/virt-top
> > > >
> > > >
> >
> > --
> > Richard Jones, Virtualization Group, Red Hat http://people.redhat.com/~rjones
> > Read my programming and virtualization blog: http://rwmj.wordpress.com
> > virt-top is 'top' for virtual machines.  Tiny program with many
> > powerful monitoring features, net stats, disk stats, logging, etc.
> > http://people.redhat.com/~rjones/virt-top
> >
> >

-- 
Richard Jones, Virtualization Group, Red Hat http://people.redhat.com/~rjones
Read my programming and virtualization blog: http://rwmj.wordpress.com
virt-builder quickly builds VMs from scratch
http://libguestfs.org/virt-builder.1.html

Re: [cocci] Very long compile times on parsing_cocci/parser_cocci_menhir.ml

[Julia Lawall]

On Thu, 2 May 2024, Richard W.M. Jones wrote:

> On Thu, May 02, 2024 at 06:39:34PM +0200, Julia Lawall wrote:
> >
> >
> > On Thu, 2 May 2024, Richard W.M. Jones wrote:
> >
> > > On Thu, May 02, 2024 at 11:44:00AM +0200, Julia Lawall wrote:
> > > >
> > > >
> > > > On Thu, 2 May 2024, Richard W.M. Jones wrote:
> > > >
> > > > > OCAMLOPT  parsing_cocci/parser_cocci_menhir.ml
> > > > >
> > > > > This single file takes a very long time to compile, about 34 minutes
> > > > > on my 2 year old Intel laptop.  It's quite large too ...
> > > > >
> > > > > 139201 ./coccinelle-1.2/parsing_cocci/parser_cocci_menhir.ml
> > > > >
> > > > > I've observed in the past that the OCaml compiler doesn't handle
> > > > > single large files well.  It seems to have some O(n^k) behaviour.
> > > > >
> > > > > Is this a known issue?
> > > >
> > > > I agree that it's slow, but I haven't seen it take 34 minutes.
> > >
> > > I just tried it on a much faster AMD desktop machine, but essentially
> > > the same OCaml and menhir, and it still takes a very long time to
> > > compile this one file: 17:35 minutes
> > >
> > > I tried it on a (different) machine with OCaml 4.11.1 and that was
> > > considerably faster, taking under a minute.
> >
> > I tried it on an Intel server and it takes a few seconds.  OCaml version
> > 5.1.1.  menhir, version 20200123
>
> Interesting, thanks Julia.  I'll poke this further to see
> if I can find out what the causative factor is.

wc -l parser_cocci_menhir.ml gives 139201 parser_cocci_menhir.ml

julia

>
> Rich.
>
> > julia
> >
> > >
> > > Rich.
> > >
> > > > The point is that it is a parser, so the code is generated.  I use OCaml
> > > > 4.13.1 or 5.0, not 5.1.1.  I have the same version of menhir as the one
> > > > you mention.
> > > >
> > > > I can try to make it smaller.
> > > >
> > > > julia
> > > >
> > > > >
> > > > > Rich.
> > > > >
> > > > > coccinelle 1.2 from tarball
> > > > > ocaml-5.1.1-4.fc40.x86_64
> > > > > ocaml-menhir-20231231-3.fc40.x86_64
> > > > >
> > > > > --
> > > > > Richard Jones, Virtualization Group, Red Hat http://people.redhat.com/~rjones
> > > > > Read my programming and virtualization blog: http://rwmj.wordpress.com
> > > > > virt-top is 'top' for virtual machines.  Tiny program with many
> > > > > powerful monitoring features, net stats, disk stats, logging, etc.
> > > > > http://people.redhat.com/~rjones/virt-top
> > > > >
> > > > >
> > >
> > > --
> > > Richard Jones, Virtualization Group, Red Hat http://people.redhat.com/~rjones
> > > Read my programming and virtualization blog: http://rwmj.wordpress.com
> > > virt-top is 'top' for virtual machines.  Tiny program with many
> > > powerful monitoring features, net stats, disk stats, logging, etc.
> > > http://people.redhat.com/~rjones/virt-top
> > >
> > >
>
> --
> Richard Jones, Virtualization Group, Red Hat http://people.redhat.com/~rjones
> Read my programming and virtualization blog: http://rwmj.wordpress.com
> virt-builder quickly builds VMs from scratch
> http://libguestfs.org/virt-builder.1.html
>
>

Re: [cocci] Very long compile times on parsing_cocci/parser_cocci_menhir.ml

[Richard W.M. Jones]

FWIW this is the flamegraph.  It's not very informative, but maybe
means a bit more to people familiar with the internals of the OCaml
compiler.

http://oirase.annexia.org/2024-05-ocamlopt.svg

Rich.

-- 
Richard Jones, Virtualization Group, Red Hat http://people.redhat.com/~rjones
Read my programming and virtualization blog: http://rwmj.wordpress.com
virt-p2v converts physical machines to virtual machines.  Boot with a
live CD or over the network (PXE) and turn machines into KVM guests.
http://libguestfs.org/virt-v2v

Re: [cocci] Very long compile times on parsing_cocci/parser_cocci_menhir.ml

[Richard W.M. Jones]

On Thu, May 02, 2024 at 06:19:30PM +0100, Richard W.M. Jones wrote:
> FWIW this is the flamegraph.  It's not very informative, but maybe

... except that 86% of the entire time is taken in these two functions
(and stuff called from there):

https://github.com/ocaml/ocaml/blob/e202de7439afaec4671da56b59dce240447ea174/middle_end/flambda/invariant_params.ml#L99-L108

whatever that means.

> means a bit more to people familiar with the internals of the OCaml
> compiler.
> 
> http://oirase.annexia.org/2024-05-ocamlopt.svg

Rich.

-- 
Richard Jones, Virtualization Group, Red Hat http://people.redhat.com/~rjones
Read my programming and virtualization blog: http://rwmj.wordpress.com
Fedora Windows cross-compiler. Compile Windows programs, test, and
build Windows installers. Over 100 libraries supported.
http://fedoraproject.org/wiki/MinGW

Re: [cocci] Very long compile times on parsing_cocci/parser_cocci_menhir.ml

[Wolfram Sang]

[-- Attachment #1: Type: text/plain, Size: 255 bytes --]


> I think there is some significant difference between OCaml 4.x and
> OCaml 5.1.1.  Can you see if updating to OCaml 5.1.1 makes it slower?

Sorry, there is no OCaml 5 for Debian yet and compiling the toolchain
itself is beyond my bandwidth currently.


[-- Attachment #2: signature.asc --]
[-- Type: application/pgp-signature, Size: 833 bytes --]

Re: [cocci] Very long compile times on parsing_cocci/parser_cocci_menhir.ml

[Julia Lawall]

On Thu, 2 May 2024, Richard W.M. Jones wrote:

> On Thu, May 02, 2024 at 06:19:30PM +0100, Richard W.M. Jones wrote:
> > FWIW this is the flamegraph.  It's not very informative, but maybe
>
> ... except that 86% of the entire time is taken in these two functions
> (and stuff called from there):
>
> https://github.com/ocaml/ocaml/blob/e202de7439afaec4671da56b59dce240447ea174/middle_end/flambda/invariant_params.ml#L99-L108
>
> whatever that means.

I don't know what it means.  Do you want me to contact the OCaml people?
Or do you want to contact them?

I made another test on another machine and still can't reproduce the
problem.

julia


>
> > means a bit more to people familiar with the internals of the OCaml
> > compiler.
> >
> > http://oirase.annexia.org/2024-05-ocamlopt.svg
>
> Rich.
>
> --
> Richard Jones, Virtualization Group, Red Hat http://people.redhat.com/~rjones
> Read my programming and virtualization blog: http://rwmj.wordpress.com
> Fedora Windows cross-compiler. Compile Windows programs, test, and
> build Windows installers. Over 100 libraries supported.
> http://fedoraproject.org/wiki/MinGW
>
>

Re: [cocci] Very long compile times on parsing_cocci/parser_cocci_menhir.ml

[Richard W.M. Jones]

On Thu, May 02, 2024 at 11:41:22PM +0200, Julia Lawall wrote:
> 
> 
> On Thu, 2 May 2024, Richard W.M. Jones wrote:
> 
> > On Thu, May 02, 2024 at 06:19:30PM +0100, Richard W.M. Jones wrote:
> > > FWIW this is the flamegraph.  It's not very informative, but maybe
> >
> > ... except that 86% of the entire time is taken in these two functions
> > (and stuff called from there):
> >
> > https://github.com/ocaml/ocaml/blob/e202de7439afaec4671da56b59dce240447ea174/middle_end/flambda/invariant_params.ml#L99-L108
> >
> > whatever that means.
> 
> I don't know what it means.  Do you want me to contact the OCaml people?
> Or do you want to contact them?
> 
> I made another test on another machine and still can't reproduce the
> problem.

Let me keep looking for a bit and see if I can find the root cause.

Rich.

> julia
> 
> 
> >
> > > means a bit more to people familiar with the internals of the OCaml
> > > compiler.
> > >
> > > http://oirase.annexia.org/2024-05-ocamlopt.svg
> >
> > Rich.
> >
> > --
> > Richard Jones, Virtualization Group, Red Hat http://people.redhat.com/~rjones
> > Read my programming and virtualization blog: http://rwmj.wordpress.com
> > Fedora Windows cross-compiler. Compile Windows programs, test, and
> > build Windows installers. Over 100 libraries supported.
> > http://fedoraproject.org/wiki/MinGW
> >
> >

-- 
Richard Jones, Virtualization Group, Red Hat http://people.redhat.com/~rjones
Read my programming and virtualization blog: http://rwmj.wordpress.com
virt-p2v converts physical machines to virtual machines.  Boot with a
live CD or over the network (PXE) and turn machines into KVM guests.
http://libguestfs.org/virt-v2v

Re: [cocci] Very long compile times on parsing_cocci/parser_cocci_menhir.ml

[Markus Elfring]

> Let me keep looking for a bit and see if I can find the root cause.

Can the analysis become easier anyhow according to questionable software run time characteristics?

Regards,
Markus

Re: [cocci] Very long compile times on parsing_cocci/parser_cocci_menhir.ml

[Richard W.M. Jones]

On Fri, May 03, 2024 at 10:00:20AM +0200, Markus Elfring wrote:
> > Let me keep looking for a bit and see if I can find the root cause.
> 
> Can the analysis become easier anyhow according to questionable software run time characteristics?

I don't understand what this means.

Rich.

> Regards,
> Markus

-- 
Richard Jones, Virtualization Group, Red Hat http://people.redhat.com/~rjones
Read my programming and virtualization blog: http://rwmj.wordpress.com
virt-p2v converts physical machines to virtual machines.  Boot with a
live CD or over the network (PXE) and turn machines into KVM guests.
http://libguestfs.org/virt-v2v

Re: [cocci] Analysing software build processes

[Markus Elfring]

>> Can the analysis become easier anyhow according to questionable software run time characteristics?
>
> I don't understand what this means.

Do you know further profiling tools?

Example:
https://bashdb.sourceforge.net/remake/


You managed to generate a flamegraph with a special software build process.
Can we benefit any more from such data representations?


Would you like to help to achieve out of source builds
for affected programs?

Regards,
Markus

Re: [cocci] Analysing software build processes

[Richard W.M. Jones]

On Fri, May 03, 2024 at 12:40:59PM +0200, Markus Elfring wrote:
> >> Can the analysis become easier anyhow according to questionable software run time characteristics?
> >
> > I don't understand what this means.
> 
> Do you know further profiling tools?
> 
> Example:
> https://bashdb.sourceforge.net/remake/

The problem seems to be entirely inside ocamlopt.opt, so I don't
understand how tracing make will help.

> You managed to generate a flamegraph with a special software build process.
> Can we benefit any more from such data representations?

There's no "special software build process".  The flamegraph was
collected by running 'perf record -a -g' on the machine while
ocamlopt.opt was compiling a single file, and converting the perf data
to a flamegraph using Brendan's tools here:
https://github.com/brendangregg/FlameGraph

Rich.

> Would you like to help to achieve out of source builds
> for affected programs?
> 
> Regards,
> Markus

-- 
Richard Jones, Virtualization Group, Red Hat http://people.redhat.com/~rjones
Read my programming and virtualization blog: http://rwmj.wordpress.com
virt-p2v converts physical machines to virtual machines.  Boot with a
live CD or over the network (PXE) and turn machines into KVM guests.
http://libguestfs.org/virt-v2v

Re: [cocci] Analysing software build processes

[Markus Elfring]

>> https://bashdb.sourceforge.net/remake/
>
> The problem seems to be entirely inside ocamlopt.opt, so I don't
> understand how tracing make will help.

I assume that you can benefit another bit from the Callgrind profile format.
https://remake.readthedocs.io/en/latest/features.html#profiling


>> You managed to generate a flamegraph with a special software build process.
>> Can we benefit any more from such data representations?
>
> There's no "special software build process".  The flamegraph was
> collected by running 'perf record -a -g' on the machine while
> ocamlopt.opt was compiling a single file, …

I guess that such data generation can be improved.

The compilation command (for which undesirable software run time characteristics
were accidentally observed) can be determined.
Another command variant can be tested then with special parameters,
can't it?


>> Would you like to help to achieve out of source builds
>> for affected programs?

You might be able to compare more OCaml compiler versions.
Would you dare to fiddle with bisecting of corresponding program variations?

Regards,
Markus

Re: [cocci] Very long compile times on parsing_cocci/parser_cocci_menhir.ml

[Richard W.M. Jones]

On Thu, May 02, 2024 at 07:02:30PM +0200, Julia Lawall wrote:
> 
> 
> On Thu, 2 May 2024, Richard W.M. Jones wrote:
> 
> > On Thu, May 02, 2024 at 06:39:34PM +0200, Julia Lawall wrote:
> > >
> > >
> > > On Thu, 2 May 2024, Richard W.M. Jones wrote:
> > >
> > > > On Thu, May 02, 2024 at 11:44:00AM +0200, Julia Lawall wrote:
> > > > >
> > > > >
> > > > > On Thu, 2 May 2024, Richard W.M. Jones wrote:
> > > > >
> > > > > > OCAMLOPT  parsing_cocci/parser_cocci_menhir.ml
> > > > > >
> > > > > > This single file takes a very long time to compile, about 34 minutes
> > > > > > on my 2 year old Intel laptop.  It's quite large too ...
> > > > > >
> > > > > > 139201 ./coccinelle-1.2/parsing_cocci/parser_cocci_menhir.ml
> > > > > >
> > > > > > I've observed in the past that the OCaml compiler doesn't handle
> > > > > > single large files well.  It seems to have some O(n^k) behaviour.
> > > > > >
> > > > > > Is this a known issue?
> > > > >
> > > > > I agree that it's slow, but I haven't seen it take 34 minutes.
> > > >
> > > > I just tried it on a much faster AMD desktop machine, but essentially
> > > > the same OCaml and menhir, and it still takes a very long time to
> > > > compile this one file: 17:35 minutes
> > > >
> > > > I tried it on a (different) machine with OCaml 4.11.1 and that was
> > > > considerably faster, taking under a minute.
> > >
> > > I tried it on an Intel server and it takes a few seconds.  OCaml version
> > > 5.1.1.  menhir, version 20200123
> >
> > Interesting, thanks Julia.  I'll poke this further to see
> > if I can find out what the causative factor is.
> 
> wc -l parser_cocci_menhir.ml gives 139201 parser_cocci_menhir.ml

It turns out this happens because Fedora enables "flambda" in the
OCaml compiler (ie. ./configure --enable-flambda).

flambda is "a series of optimisation passes provided by the native
code compilers as of OCaml 4.03" which "make it easier to write
idiomatic OCaml code without incurring performance penalties".  It's
not enabled by default, which probably explains why this problem is
not seen on Debian.

More here:
https://ocaml.org/manual/5.1/flambda.html

Anyway it's not a cocci problem!

I will discuss with Fedora if we want to keep this option enabled or
drop it.  It's not a show-stopper that Coccinelle takes a long time to
compile, just a bit annoying.  It doesn't seem to affect Coccinelle in
any other way (eg. making it faster/slower at runtime) as far as I can
tell.

Rich.

-- 
Richard Jones, Virtualization Group, Red Hat http://people.redhat.com/~rjones
Read my programming and virtualization blog: http://rwmj.wordpress.com
nbdkit - Flexible, fast NBD server with plugins
https://gitlab.com/nbdkit/nbdkit

Re: [cocci] Influence of build configuration parameter “--enable-flambda” on compilation of OCaml source files

[Markus Elfring]

> It turns out this happens because Fedora enables "flambda" in the
> OCaml compiler (ie. ./configure --enable-flambda).

How will the influence of such a build configuration parameter evolve further
for the compilation of OCaml source files?

Can it trigger adjustments for OCaml software packages?


> flambda is "a series of optimisation passes provided by the native
> code compilers as of OCaml 4.03" which "make it easier to write
> idiomatic OCaml code without incurring performance penalties".

Thanks for such background information.


> It's not enabled by default, which probably explains why this problem is
> not seen on Debian.

Will any corresponding program variations become more interesting?


> I will discuss with Fedora if we want to keep this option enabled or
> drop it.  It's not a show-stopper that Coccinelle takes a long time
> to compile, just a bit annoying.

Will corresponding software run time characteristics be checked and improved
any more?


> It doesn't seem to affect Coccinelle in any other way
> (eg. making it faster/slower at runtime) as far as I can tell.

Would anybody like to improve such comparisons in more detail?

Regards,
Markus

Re: [cocci] Very long compile times on parsing_cocci/parser_cocci_menhir.ml

[Julia Lawall]

On Fri, 3 May 2024, Richard W.M. Jones wrote:

> On Thu, May 02, 2024 at 07:02:30PM +0200, Julia Lawall wrote:
> >
> >
> > On Thu, 2 May 2024, Richard W.M. Jones wrote:
> >
> > > On Thu, May 02, 2024 at 06:39:34PM +0200, Julia Lawall wrote:
> > > >
> > > >
> > > > On Thu, 2 May 2024, Richard W.M. Jones wrote:
> > > >
> > > > > On Thu, May 02, 2024 at 11:44:00AM +0200, Julia Lawall wrote:
> > > > > >
> > > > > >
> > > > > > On Thu, 2 May 2024, Richard W.M. Jones wrote:
> > > > > >
> > > > > > > OCAMLOPT  parsing_cocci/parser_cocci_menhir.ml
> > > > > > >
> > > > > > > This single file takes a very long time to compile, about 34 minutes
> > > > > > > on my 2 year old Intel laptop.  It's quite large too ...
> > > > > > >
> > > > > > > 139201 ./coccinelle-1.2/parsing_cocci/parser_cocci_menhir.ml
> > > > > > >
> > > > > > > I've observed in the past that the OCaml compiler doesn't handle
> > > > > > > single large files well.  It seems to have some O(n^k) behaviour.
> > > > > > >
> > > > > > > Is this a known issue?
> > > > > >
> > > > > > I agree that it's slow, but I haven't seen it take 34 minutes.
> > > > >
> > > > > I just tried it on a much faster AMD desktop machine, but essentially
> > > > > the same OCaml and menhir, and it still takes a very long time to
> > > > > compile this one file: 17:35 minutes
> > > > >
> > > > > I tried it on a (different) machine with OCaml 4.11.1 and that was
> > > > > considerably faster, taking under a minute.
> > > >
> > > > I tried it on an Intel server and it takes a few seconds.  OCaml version
> > > > 5.1.1.  menhir, version 20200123
> > >
> > > Interesting, thanks Julia.  I'll poke this further to see
> > > if I can find out what the causative factor is.
> >
> > wc -l parser_cocci_menhir.ml gives 139201 parser_cocci_menhir.ml
>
> It turns out this happens because Fedora enables "flambda" in the
> OCaml compiler (ie. ./configure --enable-flambda).
>
> flambda is "a series of optimisation passes provided by the native
> code compilers as of OCaml 4.03" which "make it easier to write
> idiomatic OCaml code without incurring performance penalties".  It's
> not enabled by default, which probably explains why this problem is
> not seen on Debian.
>
> More here:
> https://ocaml.org/manual/5.1/flambda.html
>
> Anyway it's not a cocci problem!
>
> I will discuss with Fedora if we want to keep this option enabled or
> drop it.  It's not a show-stopper that Coccinelle takes a long time to
> compile, just a bit annoying.  It doesn't seem to affect Coccinelle in
> any other way (eg. making it faster/slower at runtime) as far as I can
> tell.

Thanks!

I can at least add a comment to the build instructions.

I don't know if it is worth notifying the OCaml developers.  Maybe they
know already that this can be a problem.

julia

>
> Rich.
>
> --
> Richard Jones, Virtualization Group, Red Hat http://people.redhat.com/~rjones
> Read my programming and virtualization blog: http://rwmj.wordpress.com
> nbdkit - Flexible, fast NBD server with plugins
> https://gitlab.com/nbdkit/nbdkit
>
>

Re: [cocci] Very long compile times on parsing_cocci/parser_cocci_menhir.ml

[Markus Elfring]

> I don't know if it is worth notifying the OCaml developers.  Maybe they
> know already that this can be a problem.

Would you find it worthwhile to achieve desired data processing optimisations
also by the means of a build configuration parameter like “--enable-flambda”
in a more reasonable time frame?

Are you going to experiment further with OCaml compiler variations?

Regards,
Markus

Re: [cocci] Using OCaml compiler variants?

[Markus Elfring]

> It turns out this happens because Fedora enables "flambda" in the
> OCaml compiler (ie. ./configure --enable-flambda).

I find related information interesting for further clarifications.
https://src.fedoraproject.org/rpms/ocaml/blob/f40/f/ocaml.spec#_253


> More here:
> https://ocaml.org/manual/5.1/flambda.html

A desire was indicated to benefit from compilation optimisations also
for source files of the programming language “OCaml”.
Such optimisations mean extra data processing efforts.


> I will discuss with Fedora if we want to keep this option enabled or
> drop it. …

I hope that development interests will grow for corresponding case distinctions.

The following information is also provided by the referenced software documentation:
“…
(There is no support for a single compiler that can operate in both Flambda and non-Flambda modes.)
Code compiled with Flambda cannot be linked into the same program as code compiled without Flambda.
Attempting to do this will result in a compiler error.
…”

Software builders would occasionally like to choose between standard and optimised
system configurations.
I got the impression here that only optimised builds would be directly supported
for affected OCaml software packages (in Fedora) so far.
How much were compilation durations checked during the years?

Are you looking for possibilities according to better support for other
software build variants?

Regards,
Markus

Re: [cocci] Very long compile times on parsing_cocci/parser_cocci_menhir.ml

[Richard W.M. Jones]

On Fri, May 03, 2024 at 05:50:28PM +0200, Julia Lawall wrote:
> I don't know if it is worth notifying the OCaml developers.  Maybe they
> know already that this can be a problem.

My colleague had a go at a compiler patch:

https://github.com/ocaml/ocaml/pull/13150

Rich.

-- 
Richard Jones, Virtualization Group, Red Hat http://people.redhat.com/~rjones
Read my programming and virtualization blog: http://rwmj.wordpress.com
virt-builder quickly builds VMs from scratch
http://libguestfs.org/virt-builder.1.html

Re: [cocci] Very long compile times on parsing_cocci/parser_cocci_menhir.ml

[Julia Lawall]

On Tue, 7 May 2024, Richard W.M. Jones wrote:

> On Fri, May 03, 2024 at 05:50:28PM +0200, Julia Lawall wrote:
> > I don't know if it is worth notifying the OCaml developers.  Maybe they
> > know already that this can be a problem.
>
> My colleague had a go at a compiler patch:
>
> https://github.com/ocaml/ocaml/pull/13150

That's great.  Thanks!

julia


>
> Rich.
>
> --
> Richard Jones, Virtualization Group, Red Hat http://people.redhat.com/~rjones
> Read my programming and virtualization blog: http://rwmj.wordpress.com
> virt-builder quickly builds VMs from scratch
> http://libguestfs.org/virt-builder.1.html
>
>