[GDB] Fix pretty-printer crash for sycl::item MOffset

[ianayl]

The pretty printer looks for an MOffset field in sycl::item impl without considering whether or not offset has been enabled (Offset was deprecated in SYCL 2020): Add a check to avoid printing MOffset if offset has been disabled.

[Copilot]

Pull Request Overview

This PR fixes a crash in the GDB pretty-printer for sycl::item objects when the MOffset field is not available. The fix adds exception handling to gracefully skip offset printing when the offset feature has been disabled (as deprecated in SYCL 2020).

• Adds try-catch block around offset field access and printing logic
• Prevents crashes when MOffset field is not present in sycl::item impl
• Maintains backward compatibility for items with offset enabled

[uditagarwal97]

Do you know if it's feasible to add a test for this change? I don't recall seeing tests for sycl/gdb/libsycl.so-gdb.py TBH, so not sure if it's feasible.

[ianayl]

Do you know if it's feasible to add a test for this change? I don't recall seeing tests for sycl/gdb/libsycl.so-gdb.py TBH, so not sure if it's feasible.

Looks like we don't have any testing for libsycl.so-gdb.py; the closest thing we have is https://github.com/intel/llvm/blob/sycl/sycl/test/gdb/printers.cpp, but this does not actually test the python printer script itself, but rather checks AST layouts.

Should I still go about updating said file to contain sycl::item? I had initially tried to extend this file to check that our classes contain fields that are as expected by the pretty printer, but as I actually started writing the test, I realized that since we never actually test-run libsycl.so-gdb.py, the test seems rather useless for my sakes: I assume we are trying to explicitly check if the pretty-printer has bugs in it, not whether or not SYCL RT has changes to it instead.

[uditagarwal97]

Should I still go about updating said file to contain sycl::item?

Yes

[ianayl]

I added a check in sycl/test/gdb/printers.cpp for sycl::item, checking for both MOffset and no MOffset: But as I mentioned prior, this test itself never actually runs the gdb printer 😅 it is effectively a lit test that checks SYCL constructs are as we expect them to be. I've added the test as requested, but I don't know if it does anything useful...