fix: repair master — #86, #87 and #57 merged clean and do not compile

[Kartikey1306]

eBoot master (22d8f8b) does not build. Three separate collisions, all from my #86/#87 landing alongside #57. Every one of those merges was conflict-free; none of them was correct.

This is also what is failing eos CI — Cross-compile ARM64 kernel and Full-stack integration summary on eos PRs reach this build through eos-simulation.yml.

1. include/eos_image.h asserts on a field that no longer exists

include/eos_image.h:135: error: no member named 'reserved' in 'eos_image_header_t'
include/eos_image.h:142: error: no member named 'reserved' in 'eos_image_header_t'

#87 pinned every header field, reserved[] among them. #57 then split reserved[30] into tlv_len (2) + tlv_hash (28). Both landed, and git had no reason to object — they touch different lines of the same file.

Offset 62 is already pinned by #57's tlv_len assert, so the old reserved[] offset assert is deleted rather than renamed; keeping both would pin one byte range twice. The width guard follows the 30 bytes into their two successors instead, which is what it existed to do:

EOS_IMG_STATIC_ASSERT(sizeof(tlv_len) + sizeof(tlv_hash) == 30, ...);
EOS_IMG_STATIC_ASSERT(sizeof(tlv_hash) == 28, ...);

2. core/ed25519_verify.c has two identical point_is_identity()

#86 and #57 each added one, byte-for-byte the same, plus equivalent subgroup checks under different names. Kept #57's public_key_is_valid_subgroup() — the one eos_ed25519_verify() actually calls — and #86's fuller comment, which is the one that explains why both conditions are needed and why the identity has to be excluded separately. Dropped #86's now-dead key_has_prime_order().

3. tests/unit/test_ed25519.c kept the tests and dropped their data

test_ed25519_identity_key_forgery_rejected() was defined twice. messages[] and k_low_order[8][32] were dropped while the test that iterates them survived, so the file did not compile. Restored both verbatim from the #86 branch.

Two more surfaced once it built, neither of which any compiler could have caught:

• test_ed25519_low_order_R_with_a_valid_key_is_not_a_forgery() was compiled but never called. The merge kept the function and dropped its run_ call.
• tests_run was hardcoded to 11 while 12 tests ran, so the suite failed on its own arithmetic (12/11 tests passed, exit 1). It is now incremented by the macro that runs each test, so it cannot go stale again.

Validation

	master	this PR
header-only TU	2 errors	compiles
cmake && ninja	3 errors	clean
ctest	cannot build	21/21 passed
test_ed25519	—	13/13 (was 12/11 → exit 1)
low-order forgery probe	—	0 accepted / 2048 attempts
RFC 8032 vectors	—	all still accepted

The last two matter together: the forgery probe confirms consolidating the two subgroup checks did not weaken the rejection, and the RFC vectors confirm it did not start over-rejecting. A verifier that refused everything would pass the first alone.

Worth saying plainly

This is the third time in this repo that a clean merge has produced code that does not compile, and my PRs are half of this one. The tests_run case is the one I would flag hardest: a test that is compiled but never called is invisible to the build, to ctest, and to the registration guard, which checks for files rather than call sites. eos has a live instance of the same class (test_queue_send_waiter_overflow).

🤖 Generated with Claude Code

[codecov-commenter]

⚠️ Please install the to ensure uploads and comments are reliably processed by Codecov.

Codecov Report

✅ All modified and coverable lines are covered by tests.

📢 Thoughts on this report? Let us know!

[Kartikey1306]

Closing in favour of #94, which is the same repair, opened eight hours earlier, and better in every place the two differ. My fault for not checking for an existing PR before opening this.

I verified #94 independently rather than assuming, building it from refs/pull/94/head:

cmake + ninja            clean
ctest                    21/21 passed
low-order forgery probe  0 accepted / 2048 attempts
RFC 8032 vectors         all still accepted

Where they differ, #94 wins:

	#94	this PR
TEST() blocks / run_ calls	14 / 14	13 / 13
header width asserts	sizeof(tlv_len) == 2 and sizeof(tlv_hash) == 28	sizeof(tlv_len) + sizeof(tlv_hash) == 30, plus tlv_hash == 28

Pinning the two widths individually implies the sum; pinning the sum plus one part is the weaker of the two statements. And #94 carries one more test than I did.

Both of us independently found the same three collisions (reserved[] vs the tlv_len/tlv_hash split, the duplicate point_is_identity(), and test_ed25519.c keeping tests while dropping messages[]/k_low_order[]), and both landed on tests_run++ in the macro rather than a hardcoded total. Nothing in this PR is lost by closing it.

The point from this PR's description that is worth carrying over to #94's review, since it is about the class rather than the diff: test_ed25519_low_order_R_with_a_valid_key_is_not_a_forgery() was compiled but never called — the merge kept the function and dropped its run_ call. No compiler, no linker, and no ctest run can see that, and test_cmake_test_registration.py cannot either, because it checks that each .c file has an add_executable, not that each TEST(x) has a matching run_x(). #94's tests_run++ change closes the arithmetic half of it. The symbol half is worth a follow-up guard in both repos — eos has a live instance in test_queue_send_waiter_overflow.