[Cython] question on submitting a possibly massive bug report

2022-01-25 Thread website.reader via cython-devel
Hi all:

I am not familiar with Cython, but have spent a few weeks looking at compiler 
warnings posted when the mathematical package called "sage v9.4" is compiled, 
which takes several hours to build, since hundreds of code units are invovled 
in this massive build project.

I logged 341 errors during the cythonizing part of the compile run, and found 
110 code units (C packages) which I was able to fix so that the recompile would 
have no warnings. The warnings were legitimate.

There are 4 categories of these warnings.

1. Using an unitialized variable with an unknown value
2. Comparing signed and unsigned variables
3. Discarding a const specifier to a variable upon use elsewhere in the code
4. Coercing a pointer to a variable of the wrong type (or vice versa)

I did speak to one knowledgable person about this, but my question is this

a) do I submit 341 bug reports covering all the warnings?
b) since 110 code units were affected do I file 110 bug reports for each code 
unit?
b) do I submit just one bug report for each of the 4 categories above, thus 
just 4 bug reports?
c) do I just list all the warning messages obtained from the massive build run 
so everyone can get some idea of the problems being faced?

I did look at the C code and the pyx code generating it and definitely cython 
is the origination here of these issues.

Since I am NOT yet familiar with cython from scratch, at the moment I am at a 
loss to write litte tiny programs illustrating the problem.

What do you recommend?

Randall___
cython-devel mailing list
cython-devel@python.org
https://mail.python.org/mailman/listinfo/cython-devel


Re: [Cython] question on submitting a possibly massive bug report

2022-01-25 Thread Stefan Behnel

website.reader via cython-devel schrieb am 25.01.22 um 01:09:

I am not familiar with Cython, but have spent a few weeks looking at compiler warnings 
posted when the mathematical package called "sage v9.4" is compiled, which 
takes several hours to build, since hundreds of code units are invovled in this massive 
build project.

I logged 341 errors during the cythonizing part of the compile run, and found 
110 code units (C packages) which I was able to fix so that the recompile would 
have no warnings. The warnings were legitimate.

There are 4 categories of these warnings.

1. Using an unitialized variable with an unknown value
2. Comparing signed and unsigned variables
3. Discarding a const specifier to a variable upon use elsewhere in the code
4. Coercing a pointer to a variable of the wrong type (or vice versa)

I did speak to one knowledgable person about this, but my question is this

a) do I submit 341 bug reports covering all the warnings?
b) since 110 code units were affected do I file 110 bug reports for each code 
unit?
b) do I submit just one bug report for each of the 4 categories above, thus 
just 4 bug reports?
c) do I just list all the warning messages obtained from the massive build run 
so everyone can get some idea of the problems being faced?

I did look at the C code and the pyx code generating it and definitely cython 
is the origination here of these issues.

Since I am NOT yet familiar with cython from scratch, at the moment I am at a 
loss to write litte tiny programs illustrating the problem.


Cython is a code generator, so there probably are only a few places where a 
larger bunch of issues originate from. You already grouped them by type 
(1-4), and those likely belong to one cause (or a few related causes). Just 
open one issue for each of the four. Then please list a few source code 
examples in each, together with the C code that Cython generated for them, 
and the warning that the C compiler gave you.


If we later find that not all warnings can be resolved this way, we'll see 
what we can do about the rest.


Please make sure to provide the Cython version that you are using. The 
latest release is 3.0.0a10 (and the main development goes there), although 
there is a legacy stable version series 0.29.x that most projects are still 
using and where we will continue to fix bugs for another while. But new 
reports should best target 3.0 in order to avoid chasing zombies.


Thanks,
Stefan
___
cython-devel mailing list
cython-devel@python.org
https://mail.python.org/mailman/listinfo/cython-devel