Add GOFF read/write support by dalvescb · Pull Request #969 · gimli-rs/object

dalvescb · 2026-07-29T22:46:57Z

This PR adds read and write support for GOFF

Read Support

Sections: GOFF Element Definitions (ED) and Section Definitions (SD) are mapped to sections
Symbols: A subset of symbol types are exposed in ObjectSymbolTable (LD, PR, ER), this matches what we are doing in our LLVM tooling
Non-contiguous data: Since section/segment data is split across non-contiguous TXT records, data() returns an error directing users to uncompressed_data(), which assembles the complete section data from all contributing records.
Symbol Names: GOFF uses EBCDIC encoding and supports long names via continuation records. To avoid ASCII/EBCDIC conversions and lifetime issues with non-contiguous data, name() returns an error in favor of name_parts(), which returns a vec of raw byte slices that can be assembled by the caller.
Relocations: RLD (Relocation Directory) records are parsed and exposed as relocations with full R-pointer/P-pointer information preserved in RelocationFlags::Goff. Generic RelocationKind mapping is best-effort based on relocation flags.

Write Support

Complete write implementation supporting sections, symbols, relocations, and all GOFF record types (HDR, ESD, TXT, RLD, LEN, END).
Expects raw bytes for EBCDIC-encoded symbol names (no conversion provided)

Testing

Round-trip tests in tests/round_trip/goff.rs verify write/read consistency for sections, symbols, and basic relocations.
Relocation tests in tests/round_trip/goff_relocation.rs verify relocation parsing and reconstruction.
I have read tests with sample object files and objdump output, but are not included in this PR. I'll open a separate PR in testfiles to add those first

Note: you can refer to the spec here IBM z/OS GOFF Specification

philipc · 2026-07-30T09:08:33Z

Thanks for the PR. I probably won't get to review this for a few days or more. You may have noticed I moved the GOFF support to an unstable feature in #962. My plan is to do an object release first, and then get this merged and do any followup work. Once that is done, we can remove the unstable requirement for GOFF and if it didn't require any breaking changes in the non-GOFF code then we can do a patch release.

However, if there is anywhere that you think it would be beneficial to have a breaking change in the non-GOFF code in order to accommodate GOFF then now is the time to do it. I didn't notice anything from a quick look through this PR. name_parts is currently only on GoffSymbol, and it may be beneficial to add that to ObjectSymbol too, but that won't be a breaking change.

It would be useful to have the fixtures in testfiles before I review this so that I can try running on some real files.

dalvescb · 2026-07-30T18:57:44Z

Thanks for your consideration

My plan is to do an object release first, and then get this merged and do any followup work. Once that is done, we can remove the unstable requirement for GOFF and if it didn't require any breaking changes in the non-GOFF code then we can do a patch release.

Sounds good, I'll make sure to update this PR to reflect any changes that are made if necessary inbetween this and the next release

However, if there is anywhere that you think it would be beneficial to have a breaking change in the non-GOFF code in order to accommodate GOFF then now is the time to do it.

Noted, I don't believe we'll require any breaking changes to satisfy our needs. I don't see any issue with leaving name_parts as a GoffSymbol only method for our purposes.

It would be useful to have the fixtures in testfiles before I review this so that I can try running on some real files.

I've opened a PR to add sample testfiles here gimli-rs/object-testfiles#45 . If that lands ahead of time, I'll update this PR to bump the submodule and add my read tests

philipc

Some comments from a quick read through, I haven't looked at anything in detail yet.

Also, there doesn't appear to be any testing of continuation records, either here or in the object-testfiles additions.

GOFF is still new and foreign to me, so please push back if I make any suggestions that don't make sense to you.

philipc · 2026-08-06T10:17:36Z

+        ObjectKind::Unknown
+    }
+
+    fn segments(&self) -> GoffSegmentIterator<'data, '_, R> {


The segments API in this crate is intended for executable files, so we probably shouldn't be returning any segments here.

Okay, I switched GoffSegmentIterator::next() to always return None (same approach currently used in xcoff)

The ObjectSegment implementation should be changed to unreachable!(); stubs too.

Okay, stubbed

philipc · 2026-08-11T03:00:09Z

+
+        let goffsymbol = GoffSymbol {
+            symbol_index: symbolindex,
+            esdid: esdid,


Suggested change

esdid: esdid,

esdid,

There's a few other clippy warnings too.

All clippy warnings warnings are fixed now

philipc · 2026-08-11T03:01:25Z

+}
+
+#[test]
+fn goff_relocation_multiple() {


It would be good to test continuation records for these too.

I added a new test specifically for triggering continuation records for relocations here, and a small fix in the writer

What testing have you done besides what is in this PR? I'm guessing the relocation writing code path isn't actually used for your purposes (which is fine, I just need know where things stand).

…p test

…all write fix

Curtis D'Alves added 6 commits July 29, 2026 18:16

goff: extend API and add support in common

ff72878

goff: add read support

7b83832

goff: add write support

5cbe384

goff: add round_trip test

f153f6e

goff: documentation cleanup

b0bf47a

goff: support no std and only goff

05f1f63

philipc requested changes Aug 6, 2026

View reviewed changes

Curtis D'Alves added 7 commits August 9, 2026 16:48

goff: update DWARF debug symbol detection for CDA

ed460e9

goff: return empty iterator for dynamic_symbols

abaa35e

goff: switch GoffSegmentIterator to always return None

ca481a1

goff: switch symbol name to an owned Vec<u8>

f90c070

goff: remove unused Symbol64 and RelocationItem64

6a6a505

goff: refactor names to use snake case

b561ff5

goff: add SD and ED symboltypes back into public API

d0f2363

dalvescb requested a review from philipc August 10, 2026 21:11

philipc requested changes Aug 11, 2026

View reviewed changes

philipc reviewed Aug 11, 2026

View reviewed changes

Comment thread src/common.rs Outdated

Curtis D'Alves added 11 commits August 11, 2026 00:22

goff: fix bad truncation of continuation records and extend round_tri…

e53aba0

…p test

goff: update round_trip test to include txt continuation record

05de810

goff: fix symbol iteration, now deterministic

2937065

goff: fix bad XPLINK flag mask

22f8d38

goff: fix behavioral attribute flag masks

f9e7e52

goff: fix masks for GOFF_BINDING_SCOPE

058b077

goff: switch GoffFile::symbols from HashMap to Vec

b70056c

goff: stub out segments methods (and run fmt)

37a03e6

goff: remove segments flags

55f299a

goff: fix clippy suggestions

06a6de6

goff: add round_trip test for relocations (with cont records), and sm…

a361bce

…all write fix

Uh oh!

Conversation

dalvescb commented Jul 29, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

philipc commented Jul 30, 2026

Uh oh!

dalvescb commented Jul 30, 2026

Uh oh!

philipc 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!

philipc Aug 11, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

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!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

dalvescb commented Jul 29, 2026 •

edited

Loading

philipc Aug 11, 2026 •

edited

Loading