You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
This is really an issue in Hera, but it seems to affect GUDHI very directly. I have higher hopes of seeing a response here.
Until some weeks ago, the GUDHI Debian packages accidentally built the (C++ part of the) Python bindings with -DNDEBUG, thus disabling all assertions. After that was changed, I've been seeing the test suite fail. That failure seems to stem from this assertion in Hera. I'm not well-versed in the Hera code, but the assertion seems wrong to me. Won't every point in the loop just above the assertion pass the if_normal test in case of entirely-off-diagonal persistence diagrams? Then the assertion trivially fails.
This in turn causes problems for GUDHI – for example, this part of a test fails on the assertion failing. This might not easily get caught if one, like me, accidentally builds with -DNDEBUG.
Am I missing something?
The text was updated successfully, but these errors were encountered:
Thanks for the report, we do indeed have something to fix in gudhi. It also reminded us that we wanted to run (part of) the CI with assertions enabled, Vincent is looking at it in #1163.
Until some weeks ago, the GUDHI Debian packages accidentally built the (C++ part of the) Python bindings with -DNDEBUG, thus disabling all assertions.
What do you mean "accidentally"? Building the debian package with -DNDEBUG looks normal to me.
Thanks for the report, we do indeed have something to fix in gudhi.
Thanks for taking a look!
It also reminded us that we wanted to run (part of) the CI with
assertions enabled, Vincent is looking at it in #1163.
Sounds good.
Until some weeks ago, the GUDHI Debian packages accidentally built the (C++ part of the) Python
bindings with -DNDEBUG, thus disabling all assertions.
What do you mean "accidentally"? Building the debian package with
-DNDEBUG looks normal to me.
Debian desires to build our packages with assertions enabled, and so we
avoid `-DNDEBUG`. It still sneaks in a lot of cases where the authors of
upstream code have relied on it also for debugging (rather than toggling
assertions), or where (as in this case) an interaction between multiple
build system for multiple languages causes it to be enabled without the
maintainer (me) noticing.
This is really an issue in Hera, but it seems to affect GUDHI very directly. I have higher hopes of seeing a response here.
Until some weeks ago, the GUDHI Debian packages accidentally built the (C++ part of the) Python bindings with
-DNDEBUG
, thus disabling all assertions. After that was changed, I've been seeing the test suite fail. That failure seems to stem from this assertion in Hera. I'm not well-versed in the Hera code, but the assertion seems wrong to me. Won't every point in the loop just above the assertion pass theif_normal
test in case of entirely-off-diagonal persistence diagrams? Then the assertion trivially fails.This in turn causes problems for GUDHI – for example, this part of a test fails on the assertion failing. This might not easily get caught if one, like me, accidentally builds with
-DNDEBUG
.Am I missing something?
The text was updated successfully, but these errors were encountered: