[BUG] invalid tag name with iterators in --gdb mode

[tornaria]

Describe the bug

$ cat example.pyx 
def use_an_iterator(iterator):
    return (x for x in iterator)
$ cython3 --version
Cython version 3.0.0
$ cython3 -3 example.pyx 
$ cython3 -3 --gdb example.pyx 

Error compiling Cython file:
------------------------------------------------------------
...
def use_an_iterator(iterator):
^
------------------------------------------------------------

example.pyx:1:0: Compiler crash in DebugTransform


Compiler crash traceback from this point on:
  File "Cython/Compiler/Visitor.py", line 182, in Cython.Compiler.Visitor.TreeVisitor._visit
  File "/usr/lib/python3.11/site-packages/Cython/Compiler/ParseTreeTransforms.py", line 4033, in visit_ModuleNode
    self.visit_FuncDefNode(nested_funcdef)
  File "/usr/lib/python3.11/site-packages/Cython/Compiler/ParseTreeTransforms.py", line 4093, in visit_FuncDefNode
    self.tb.start(arg.name)
  File "/usr/lib/python3.11/site-packages/Cython/Debugger/DebugWriter.py", line 42, in start
    self.tb.start(name, attrs or {})
  File "src/lxml/saxparser.pxi", line 841, in lxml.etree.TreeBuilder.start
  File "src/lxml/saxparser.pxi", line 769, in lxml.etree.TreeBuilder._handleSaxStart
  File "src/lxml/apihelpers.pxi", line 179, in lxml.etree._makeSubElement
  File "src/lxml/apihelpers.pxi", line 1754, in lxml.etree._tagValidOrRaise
ValueError: Invalid tag name '.0'

Code to reproduce the behaviour:

def use_an_iterator(iterator):
    return (x for x in iterator)

Expected behaviour

No response

OS

linux

Python version

3.11.4

Cython version

No response

Additional context

This happens only if python lxml package is installed.

I reduced it to the simple function above, but this happens a lot in sagemath source code.

[da-woods]

So the origin of this is: I've used function arguments with names like .0 and .1 as a way of passing arguments into generator expressions. This is by analogy with Python where

>>> [locals() for _ in range(1)]
[{'.0': <range_iterator object at 0x7f9a4732c150>, '_': 0}]

(exact behaviour likely somewhat version dependent).

It looks like these don't play well with the debug output writer.

It's probably fairly simple to fix either by renaming the arguments or by excluding them from the debug output. I don't know much about the --gdb so I don't really know what the appropriate solution is. But one of those solutions should work.

[h-vetinari]

This issue is also affecting pyarrow, FWIW

[scoder]

We defined the XML format ourselves, I think, but changing it now brings in compatibility problems.

I think excluding these arguments is probably ok for now.

[oleksandr-pavlyk]

I would like to look into fixing this issue, as it blocks building debug builds for certain projects based on scikit-build which passes --gdb argument to cython automatically.

[da-woods]

@oleksandr-pavlyk I've assigned it to you. Hopefully it should be reasonably simple, but less us know how you get on.