[flang-commits] [flang] [flang][Parser][NFC] Don't build a discarded list while scanning identifiers (PR #219330)

[Eugene Epshteyn via flang-commits]

https://github.com/eugeneepshteyn approved this pull request.

LGTM.

Two non-blocking notes:

- Description nit: the list costs one node per identifier character *after the first* (the leading `nonDigitIdChar` is outside `many()`).
- The same discarded-list pattern exists in a few sibling spots — e.g. `some(definedOpNameChar)` in `DefinedOpName` (`expr-parsers.cpp`), the `!(some(letter) >> "."_ch)` REAL-literal lookahead, the directive-sentinel `some(letter)`, and the `many(...)` in `progUnitEndStmtErrorRecovery` — those would need `x >> skipMany(x)` (or a new `skipSome`) since `some()` has no skip variant. Fine as separate follow-ups; no need to grow this PR.

AI agent verified the performance claim:

**The allocation claim reproduces.** On a generated identifier-heavy file:

```sh
python3 -c '
n = 3000
print("program p10"); print("  implicit none")
for i in range(n): print(f"  integer :: some_quite_long_variable_name_number_{i:04d}")
print("  some_quite_long_variable_name_number_0000 = 1")
for i in range(1, n):
    print(f"  some_quite_long_variable_name_number_{i:04d} = some_quite_long_variable_name_number_{i-1:04d} + {i}")
print(f"  print *, some_quite_long_variable_name_number_{n-1:04d}"); print("end program p10")
' > p10.f90
valgrind flang -fc1 -fsyntax-only p10.f90
```

```console
before: total heap usage: 2,140,700 allocs, 93,564,066 bytes allocated
after:  total heap usage:   700,308 allocs, 58,994,632 bytes allocated
```

— this one-liner removes 67% of *all* heap allocations for that compile (−18.9% instructions:u with `-fc1 -fsyntax-only`), consistent in direction with your whole-compile −0.88%.

https://github.com/llvm/llvm-project/pull/219330