Dear Federico
Thank you for your answer, and my deepest apologies for the almost one-year delay to answer…
Reproducing the error is hard in the sense that it is presumably
linked to garbage collector, and references deleted while they
should not be.
Mikael Bisgaard Dahlsen-Jensen (cc-ed) did a great job lately by trying to isolate the issue on small examples, with valgrind, and prepared examples and explanations:
Most explanations are in the README.md file. While some errors are presumably easy to fix (some used exceptions are not registered in the PPL interface), the main issue seems related to using already-free'd memory.
This bug is terribly annoying, as it prevents us to use newer
algorithms in IMITATOR for anything but extremely small models.
Please feel free to contact us for more details, and notably Mikael.
Many thanks!
Étienne André Nantes Université https://lipn.univ-paris13.fr/~andre/
Dear Étienne,
thank you for contacting us directly.
The ppl-devel mailing list has been relocated under a different domain
(bugseng.com), we did not receive you emails.
A notice informing of the relocation has been sent to all mailing list
subscribers on Jan 22. Subject was: "PPL-devel mailing list relocation".
Please, see the new PPL-devel home at:
https://support.bugseng.com/cgi-bin/mailman/listinfo/ppl-devel
Thank you for the bug report, we will look at it as soon as possible.
Can you please share with us a testcase to reproduce the error?
Do you have any additional information that might be helpful?
Thank you, regards
Federico
On 08/11/24 14:00, Étienne André wrote:
Dear PPL developers,
I received a "mailer daemon" in reply to my former email related to a likely bug in the OCaml interface, so I take the liberty to write another email (see below).
Many thanks for your answer.
Best,
Étienne André
Université Sorbonne Paris Nord
https://lipn.univ-paris13.fr/~andre/
Le 22/10/2024 à 15:09, Étienne André a écrit :
Dear PPL developers,
We didn't have news since March, but we tried to investigate the problem, and it looks now rather likely to us that the bug mentioned in my first email /does/ come from a bug in PPL, more precisely from the OCaml binding. It is likely that some "free" operations on pointers in the binding are unnecessary, and cause the garbage collector to access some already destroyed objects, leading to segmentation faults.
My colleagues are trying to fix the binding so as to remove the "free" operations, and let the OCaml garbage collector do its job naturally.
We still want to point out this issue, and we do encourage you in investigating this matter.
Best,
Étienne André
Université Sorbonne Paris Nord
https://lipn.univ-paris13.fr/~andre/
Em sex., 31 de mai. de 2024 às 13:34, Étienne André <Andre.Etienne@lipn13.fr> escreveu:
Dear PPL developers,
I wonder whether you had received our email, and might have an
idea of why we encounter an issue of corrupted memory with
Pointset_Powerset.
Thank you!
Best,
Étienne André
Université Sorbonne Paris Nord
https://lipn.univ-paris13.fr/~andre/
Le 18/03/2024 à 11:09, Étienne André a écrit :
Dear PPL developers,
We got an issue in developing IMITATOR with the PPL OCaml
binding, most probably not related to a PPL bug (even if we
don't fully exclude it either).
Our goal is to perform the disjunction of two Pointset_Powerset,
i.e., union_assign p1 p2, where p1 gets as the result the union
of p1 and p2.
I don't think there is a function to directly do this in PPL? If
there is, which one should we use?
Our algorithm is as follows:
1. get all disjuncts of p2 (using an iterator)
2. for each disjunct, add it to p1 using the PPL OCaml binding
function ppl_Pointset_Powerset_NNC_Polyhedron_add_disjunct
Is that the right way to do?
We got dimensions exceptions, both very random (changing the
input file name yields a different exception!) and with
completely funny dimensions (such as 208 or 416) for a system
that has only 2 dimensions, and/or segmentation faults/core dump
which might come from a corrupted memory somewhere.
We use OCaml 4.14.1 with (presumably) PPL 1:1.2-8.1build1.
Thanks!
Best,
-- Étienne André
Université Sorbonne Paris Nord
https://lipn.univ-paris13.fr/~andre/
--
att. Tomaz Gomes