odb: guard null-pointer deref in getGCellTileSize()'s layer lookup

[dhgaddy]

Summary

Symptom

dbBlock::getGCellTileSize() segfaults instead of raising the intended ODB-0358 error, on a technology whose leading routing layers are backside metals (e.g. a BSPDN stack):

Program received signal SIGSEGV, Segmentation fault.
0x0000000000dddfd3 in odb::dbTechLayer::getName (this=0x0) at ./src/odb/src/db/dbTechLayer.cpp:1313
1313      return layer->name_;
#0  odb::dbTechLayer::getName (this=0x0) at ./src/odb/src/db/dbTechLayer.cpp:1313
#1  odb::dbBlock::getGCellTileSize()::$_0::operator()(int) const (layer_idx=2) at ./src/odb/src/db/dbBlock.cpp:2545
#2  odb::dbBlock::getGCellTileSize (this=0x7ffff779c6f8) at ./src/odb/src/db/dbBlock.cpp:2564

Root cause

getAverageTrackSpacing(), a lambda inside getGCellTileSize(), finds the Nth frontside (non-backside) ROUTING layer by counting layers while skipping any with dbTechLayer::isBackside() set. The baseline call site always requests the 2nd, 3rd, and 4th frontside layer once the block's max routing layer (raw, backside-inclusive numbering) is >= 4, regardless of how many frontside layers actually exist below it. When a technology has fewer than N frontside layers before that point, the search loop never assigns tech_layer, leaving it null.

The existing null check on the resulting track_grid then built its error message by unconditionally calling tech_layer->getName() -- dereferencing the same pointer the check just proved could be null, crashing on the error-reporting path itself instead of raising ODB-0358.

Fix

Guard the error message against a null tech_layer, reporting which frontside layer ordinal was requested and how many actually exist instead of dereferencing it.

Testing

New regression src/odb/test/cpp/TestGCellTileSize.cpp, registered in both CMake and Bazel. It constructs a tech with 4 backside ROUTING layers ahead of a single frontside layer (M1) and a max routing layer past the early-return threshold, then asserts getGCellTileSize() throws std::runtime_error (the documented behavior of Logger::error()) instead of crashing. Confirmed by reverting just the fix hunk: the test segfaults against master, passes with this change.

bazel test //src/odb/test/cpp:TestGCellTileSize and bazel test //:dup_id_test both pass. clang-format --dry-run -Werror is clean on both changed files.

Type of Change

• Bug fix

Impact

No crash (ODB-0358 error instead) when dbBlock::getGCellTileSize() is called on a technology with fewer frontside routing layers than the lookup expects.

Verification

• I have verified that the local build succeeds (./etc/Build.sh).
• I have run the relevant tests and they pass.
• My code follows the repository's formatting guidelines.
• I have included tests to prevent regressions.
• I have signed my commits (DCO).

Related Issues

#11263

[gemini-code-assist]

Code Review

This pull request fixes a potential null-pointer dereference in dbBlock::getGCellTileSize() when a technology contains fewer frontside routing layers than expected (for example, when backside routing layers precede frontside layers). It safely handles a null tech_layer by formatting an informative error message instead of crashing. Additionally, a new unit test TestGCellTileSize has been added to verify this behavior and prevent regressions. I have no further feedback to provide on these changes.

[QuantamHD]

If the tech layer isn't found we should error out before we call findTrackGrid. Please break up the error message into two separate errors

[osamahammad21]

Thanks @dhgaddy