[PATCH 00/17] Rewrite .debug_names reader and writer

[Tom Tromey <tom@tromey.com>]

This series rewrites gdb's .debug_names reader and writer.

Currently, gdb can generate a .debug_names index, but the contents are
very wrong -- they use fully-qualified names, which is not at all what
is envisioned in the DWARF spec.

After this series, the section contents will be much closer to DWARF.
However, it's not really useful for gdb to follow DWARF exactly.

First, the DWARF spec has a few bugs affecting the generation of the
index.  GDB fixes these -- you won't see these fixes in this series,
though, as they were all handled by the new DWARF scanner.

Second, gdb likes to have some information not readily available in
the DWARF-specified contents.  You can see the new DW_IDX_* constants
patch for a list.

I've regression tested this using cc-with-debug-names on x86-64
Fedora 38.

It's possible that there are regressions with cc-with-debug-names at
some intermediate points in the series -- in particular I think it
would happen after the reader is rewritten but before the writer patch
is applied.  I didn't test this situation, since I don't think it is
very important.  It could be worked around by combining the two
patches, but I thought that would make review much more difficult.

This is based on my earlier series to read DWARF in the background.
This was done because (1) that's going to land eventually anyway and
(2) that series generalizes the cooked index code somewhat, making it
a nicer starting point.

---
Tom Tromey (17):
      Refactor 'maint set dwarf synchronous' handling
      Refactor quick-function installation in DWARF reader
      Remove IS_ENUM_CLASS from cooked_index_flag
      Add some new DW_IDX_* values
      Document GDB extensions to DWARF .debug_names
      Add language to cooked_index_entry
      Move cooked_index_functions to cooked-index.h
      Do not write the index cache from an index
      Change cooked_index_worker to abstract base class
      Remove cooked_index_worker::start_reading
      Empty hash table fix in .debug_names reader
      Fix dw2-zero-range.exp when an index is in use
      Explicitly expand CUs in dw2-inline-with-lexical-scope.exp
      Remove some .debug_names tests
      Rewrite .debug_names reader
      Export dwarf5_augmentation
      Rewrite .debug_names writer

 gdb/doc/gdb.texinfo                                |  39 +
 gdb/dwarf2/cooked-index.c                          | 162 +++-
 gdb/dwarf2/cooked-index.h                          | 184 +++-
 gdb/dwarf2/index-write.c                           | 401 ++++-----
 gdb/dwarf2/mapped-index.h                          |   4 +-
 gdb/dwarf2/read-debug-names.c                      | 949 ++++++++-------------
 gdb/dwarf2/read-debug-names.h                      |   2 +
 gdb/dwarf2/read.c                                  | 337 ++------
 gdb/testsuite/gdb.dwarf2/clang-debug-names-2-foo.c |  22 -
 gdb/testsuite/gdb.dwarf2/clang-debug-names-2.c     |  27 -
 gdb/testsuite/gdb.dwarf2/clang-debug-names-2.exp   |  42 -
 gdb/testsuite/gdb.dwarf2/clang-debug-names.c       |  25 -
 gdb/testsuite/gdb.dwarf2/clang-debug-names.exp     |  42 -
 gdb/testsuite/gdb.dwarf2/clang-debug-names.exp.tcl | 121 ---
 .../gdb.dwarf2/dw2-inline-with-lexical-scope.exp   |   4 +
 gdb/testsuite/gdb.dwarf2/dw2-zero-range.exp        |  10 +-
 include/dwarf2.def                                 |   9 +
 17 files changed, 948 insertions(+), 1432 deletions(-)
---
base-commit: 98eea77e9ce69ba00d5f93dd2264f1d1971fd86f
change-id: 20231210-debug-names-fix-cfbaf8761ecc

Best regards,
-- 
Tom Tromey <tom@tromey.com>