feat: Support GDB index by levietduc0712 · Pull Request #1605 · wild-linker/wild

levietduc0712 · 2026-02-27T05:07:16Z

This pull request adds support for generating a .gdb_index section in ELF binaries, which is used to speed up debugging with GDB. The implementation introduces a new command-line flag (--gdb-index and --no-gdb-index), updates the layout and writer logic to allocate and fill in the section, and integrates the section into the ELF output as an optional part. It also updates relevant tests and internal data structures to accommodate the new section.

Resolve #811

…o sections

…and implement gdb-index tests

…writer feat: define GDB_INDEX_SECTION_NAME constant and update section definitions

… gdb-index test configuration

…figuration

davidlattimore

Nice! Have you tried loading a binary with the gdb index into gdb to confirm that it's happy with it?

Any ideas about any other testing - automated or otherwise we could do? Checking that the section exists is a start, but it leaves a lot of scope for things being not quite right.

davidlattimore · 2026-02-27T10:43:11Z

-                self.rela_dyn_relative.len() as u64 * elf::RELA_ENTRY_SIZE,
-                *mem_sizes.get(part_id::RELA_DYN_RELATIVE),
-            ));
+            let remaining = self.rela_dyn_relative.len() as u64 * elf::RELA_ENTRY_SIZE;


Is this related to this change? I'm also curious as to the motivation for this bit of the change. It looks like you've inlined the function excessive_allocation into several of its usage sites.

It’s not required for the .gdb_index work it was just a small refactor to make the error paths explicit, which I forgot to remove before committing. I’ve restored the original code.

davidlattimore · 2026-02-27T10:50:03Z

        tracing::debug!(loaded_debug_section = %self.object.section_display_name(section_index),);
        common.section_loaded(part_id, header, section);
+
+        if self


Currently we read the section names only during section resolution. Doing an extra read of the section name can cause a noticeable slowdown. So I think we likely need to find a different way to do this.

davidlattimore · 2026-02-27T10:57:41Z

+
+    for (i, &(cu_offset, cu_length)) in cu_entries.iter().enumerate() {
+        let off = cu_list_offset + i * CU_ENTRY_SIZE;
+        buf[off..off + 8].copy_from_slice(&cu_offset.to_le_bytes());


Would it work to simplify some of this code by using zerocopy?

…ELF writer and layout

levietduc0712 · 2026-02-27T16:18:41Z

Nice! Have you tried loading a binary with the gdb index into gdb to confirm that it's happy with it?

Yes. I tested that explicitly.

I built a small test binary with -Wl,--gdb-index and then loaded it into GDB in batch mode to check for any structural issues. GDB accepted the binary without reporting any corruption, errors, or warnings.

In addition, I compared the generated .gdb_index section against an LLD-linked reference build. The section exists, the size matches, and the hex dump is identical.

So from GDB’s perspective, it appears to be fully valid and functional.

levietduc0712 · 2026-02-27T16:42:45Z

Any ideas about any other testing - automated or otherwise we could do? Checking that the section exists is a start, but it leaves a lot of scope for things being not quite right.

I’ve added a ValidateGdbIndex directive that parses the raw .gdb_index and verifies: version = 7, header offsets are ordered and in-bounds, CU entries are non-overlapping and fully cover .debug_info, and address area entries fall within .text. I also added ExpectGdbIndexCuCount: 3 to assert the exact CU count. Additionally, I loaded the binary in GDB and confirmed info sources lists files from all three input objects. Would this satisfy your request?

…integration tests

davidlattimore

The extra testing is great. Thanks!

lapla-cogito · 2026-02-28T01:52:09Z

+                        {
+                            cu_entries.push(elf::GdbIndexCuEntry {
+                                cu_offset: address - debug_info_start,
+                                cu_length: section.capacity(),


This seems to include alignment padding, making it inappropriate as a proper CU length.

…ntry validation

lapla-cogito

While I haven't thoroughly reviewed the updated patch yet (today is a bit busy, so it may take some time), I'll point out one observation immediately: Wild runs mold's test suite in CI. However, some tests fail under Wild's current implementation, so they are excluded from the skip list and ignored during runs. These skipped tests are subsequently verified to ensure they still fail in CI (though false positives occasionally occur, which is why the workflow is configured with continue-on-error: true).

Through 2a20ff7 in this PR, all tests in the GDB index category appear to pass (see this run), but the latest commit seems to cause many of these tests to fail again.

…omputation

levietduc0712 · 2026-02-28T04:09:31Z

Through 2a20ff7 in this PR, all tests in the GDB index category appear to pass (see this run), but the latest commit seems to cause many of these tests to fail again.

I fixed it. Check this one

davidlattimore · 2026-02-28T05:17:52Z

Can you delete the lines from mold_skip_tests.toml for the tests that now pass?

Also, I was thinking that it might be good to enable TestUpdateInPlace for your new test, just to make sure that all the bytes of the new section are getting written.

… source configuration

…argument parser updates

…table-compressed-debug-info test

levietduc0712 · 2026-02-28T13:20:52Z

This feature appears to be more complex than I initially estimated. I need to add gdb_index.rs file to implements hashing, hash-table sizing, parsing of .debug_gnu_pubnames/.debug_gnu_pubtypes, and writing of the symbol table + constant-pool required by the GDB index format see this one.

marxin

Really cool PR, good work. Am curious if we want to support the latest version9 (could be a subsequent PR)? Thank you.

marxin · 2026-02-28T16:42:22Z

+    for &c in name {
+        r = r
+            .wrapping_mul(67)
+            .wrapping_add(u32::from(c))


Looking at the formula: The formula is r = r * 67 + tolower (c) - 113.
Don't we miss the tolower?

marxin · 2026-02-28T16:44:06Z

+        return 0;
+    }
+    let min_slots = num_symbols * 4 / 3 + 1;
+    let mut slots = 4usize;


What about slot = (num_symbols * 4 / 3 + 1).next_multiple_of(4)?

marxin · 2026-02-28T16:47:59Z

+                .map(|obj| {
+                    let pubnames = obj
+                        .object
+                        .section_by_name(".debug_gnu_pubnames")


Please add new constants to https://github.com/davidlattimore/wild/blob/56f438d1e3c1ddbd9ff85cb0681f25aeec1ab249/linker-utils/src/elf.rs#L687

marxin · 2026-02-28T16:48:14Z

+
+        if resources
+            .output_sections
+            .custom_name_to_id(SectionName(b".debug_info"))


Similarly here.

marxin · 2026-02-28T16:49:50Z

+            );
+        }
+
+        let version = u32::from_le_bytes(data[0..4].try_into().unwrap());


Can't we directly read the GdbIndexHeader here (using the zerocopy library)?

marxin · 2026-02-28T16:50:15Z

+        let constant_pool_offset = u32::from_le_bytes(data[20..24].try_into().unwrap()) as usize;
+
+        // Check version.
+        if version != 7 {


Maybe we should declare a constant somwhere for the currently supported version.

marxin · 2026-02-28T16:51:52Z

+            .collect();
+
+        let addr_entry_count = address_area_size / 20;
+        for i in 0..addr_entry_count {


Maybe use chunks_exact?

levietduc0712 · 2026-03-01T00:38:43Z

Really cool PR, good work. Am curious if we want to support the latest version9 (could be a subsequent PR)? Thank you.

Thanks! I’ll look into supporting the latest version (9) as well.

levietduc0712 added 5 commits February 25, 2026 23:05

add initial support for --gdb-index option

52629a4

refactor: Enhance .gdb_index generation to support multiple debug_inf…

7bc9f6b

…o sections

Merge remote-tracking branch 'origin/main' into test_gdb_index

ef09aff

feat: add support for verifying section existence in output binaries …

631eae1

…and implement gdb-index tests

refactor: clean up the code

a69147b

levietduc0712 marked this pull request as draft February 27, 2026 05:07

levietduc0712 added 3 commits February 27, 2026 12:18

refactor: improve error messages and clean up code formatting in ELF …

053bf78

…writer feat: define GDB_INDEX_SECTION_NAME constant and update section definitions

refactor: format code for better readability in ELF writer and update…

83e9d4b

… gdb-index test configuration

refactor: update gdb index allocation condition and clean up test con…

5acf6df

…figuration

levietduc0712 marked this pull request as ready for review February 27, 2026 05:48

lapla-cogito self-requested a review February 27, 2026 06:31

davidlattimore reviewed Feb 27, 2026

View reviewed changes

feat: implement GDB index structures and update related functions in …

0ad734b

…ELF writer and layout

levietduc0712 force-pushed the test_gdb_index branch from 54feb16 to 8f23f3c Compare February 27, 2026 16:50

feat: add validation for .gdb_index section and expected CU count in …

2a20ff7

…integration tests

levietduc0712 force-pushed the test_gdb_index branch from 8f23f3c to 2a20ff7 Compare February 27, 2026 16:56

davidlattimore approved these changes Feb 27, 2026

View reviewed changes

lapla-cogito reviewed Feb 28, 2026

View reviewed changes

Comment thread libwild/src/elf_writer.rs

feat: enhance GDB index handling with allocation checks and address e…

3922430

…ntry validation

levietduc0712 requested a review from lapla-cogito February 28, 2026 02:48

lapla-cogito changed the title ~~feat: Add support for --gdb-index option and enhance .gdb_index generation~~ feat: Support GDB index Feb 28, 2026

lapla-cogito reviewed Feb 28, 2026

View reviewed changes

feat: calculate total address entries for GDB index and update size c…

3c5bf71

…omputation

levietduc0712 added 2 commits February 28, 2026 18:15

feat: remove GDB index tests from skipped groups and update gdb-index…

6f2ee3d

… source configuration

feat: implement GDB index handling with symbol table computation and …

94f928b

…argument parser updates

levietduc0712 marked this pull request as draft February 28, 2026 12:17

levietduc0712 added 2 commits February 28, 2026 19:35

fix: correct constant_pool_offset in gdb_index header and skip reloca…

cad47f0

…table-compressed-debug-info test

refactor: simplify is_static computation in attrs_to_cu_entry function

cd1796e

levietduc0712 marked this pull request as ready for review February 28, 2026 13:13

marxin reviewed Feb 28, 2026

View reviewed changes

levietduc0712 closed this Mar 1, 2026

Uh oh!

Conversation

levietduc0712 commented Feb 27, 2026 • edited by lapla-cogito Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

davidlattimore left a comment

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

levietduc0712 commented Feb 27, 2026

Uh oh!

levietduc0712 commented Feb 27, 2026

Uh oh!

davidlattimore left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Uh oh!

Uh oh!

lapla-cogito left a comment

Choose a reason for hiding this comment

Uh oh!

levietduc0712 commented Feb 28, 2026

Uh oh!

davidlattimore commented Feb 28, 2026

Uh oh!

levietduc0712 commented Feb 28, 2026

Uh oh!

marxin left a comment

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

levietduc0712 commented Mar 1, 2026

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

4 participants

levietduc0712 commented Feb 27, 2026 •

edited by lapla-cogito

Loading