Fix: memory output drainage and add regression test for #2469

[MakerYuichi]

Description

Fixes an issue where rnp_output_memory_get_buf() unexpectedly fails with RNP_ERROR_BAD_PARAMETERS when draining a legitimately empty memory output (such as a zero-length literal packet or an operation that produced no data).

A memory destination that hasn't written any bytes does not allocate an internal buffer, causing mem_dest_get_memory() to return NULL. This PR introduces a check to safely return RNP_SUCCESS with *len = 0 and *buf = NULL if *len == 0 on a valid PGP_STREAM_MEMORY type stream, allowing callers to uniformly drain empty memory outputs.

Fixes #2469

PR Checklist

• Branch is rebased on the latest main.
• clang-format clean (Note: clang-format wasn't available locally, relying on CI check).
• All tests pass on at least one backend (Verified 288/288 passing via Botan/macOS local build).
• Added a dedicated regression test that fails without this change (src/tests/issues/2469.cpp).
• Description explains the why and links the relevant issue.
• No observable API changes to public headers in include/rnp/rnp.h.

[ronaldtse]

Thank you @MakerYuichi ! We will be looking at this next.

[ronaldtse]

Validated end to end against a main-identical tree:

• Bug confirmed: rnp_output_to_memory never preallocates (init_mem_dest(dst, NULL, …) → param->memory = NULL), the first allocation happens in mem_dst_write, so draining a valid-but-empty memory output hit if (!*buf) return RNP_ERROR_BAD_PARAMETERS. The regression test fails on main exactly as reported in rnp_output_memory_get_buf() fails for a memory output with zero bytes written #2469 (verified locally: 0x10000002), passes with the fix.
• Fix is correct: the !*buf && *len guard keeps the defensive error for the impossible NULL-with-bytes case, the explicit PGP_STREAM_MEMORY check makes the wrong-type rejection intentional instead of incidental, and do_copy && *len avoids a pointless malloc(0).
• No overlap: no open PR modifies this function (checked all of them; dump: walk multiple armored messages in --list-packets (#2036) #2429/ffi: entropy encoding + backup archive building blocks #2435 only call it).
• Pushed a small follow-up commit (via maintainer edits): extended the test with the do_copy=true-empty and non-memory-output-rejection cases, and cleaned up style (EOF newline, trailing whitespace, unused includes, stray blank-line hunks in ffi.cpp).

This unblocks uniform empty-output draining, which e.g. rnp_backup_archive_load's plaintext drain (#2435) also relies on.

[ronaldtse]

Thank you @MakerYuichi — clear bug report shape (issue + minimal fix + regression test), exactly how we like them.

Validated end to end against a main-identical tree: the regression test fails on current main with RNP_ERROR_BAD_PARAMETERS exactly as #2469 describes (lazy allocation means an empty memory output has no buffer), and passes with this fix. All three changed paths check out: the !*buf && *len guard keeps the defensive error for the impossible NULL-with-bytes case, the explicit PGP_STREAM_MEMORY check makes wrong-type rejection intentional, and do_copy && *len avoids a pointless malloc(0). No other open PR touches this function. I pushed a small follow-up commit extending the test to the do_copy=true-empty and non-memory-output cases and cleaning up style.

[ronaldtse]

@ni4 this one needs your approval (1 of 2 — I've approved): #2472

Small, well-scoped fix by @MakerYuichi for #2469: rnp_output_memory_get_buf() returned BAD_PARAMETERS when draining a valid-but-empty memory output, because the buffer allocates lazily on first write. Now returns SUCCESS with (NULL, 0). Verified against main locally (regression test fails pre-fix, passes after), and it unblocks uniform empty-output draining — e.g. the backup-archive plaintext drain in #2435. CI green.

[ni4]

LGTM, thanks! But let's wait till non-approved workflows are finished and green.

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 85.40%. Comparing base (7853ae9) to head (e9412b8).

Additional details and impacted files

@@           Coverage Diff           @@
##             main    #2472   +/-   ##
=======================================
  Coverage   85.40%   85.40%           
=======================================
  Files         126      126           
  Lines       22964    22966    +2     
=======================================
+ Hits        19612    19614    +2     
  Misses       3352     3352           

☔ View full report in Codecov by Harness.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.

[ronaldtse]

The windows-11-arm leg fails with SHLWAPI_LIBRARY NOTFOUND — that's a pre-existing ARM64 SDK-discovery bug in main, not caused by this PR (which only touches rnp_output_memory_get_buf and test files). It's fixed by #2471 (SHLWAPI linked by name instead of the broken GetUMWindowsSDKLibraryDir probing), which also makes all four windows preview legs blocking for the first time. Once #2471 merges, this leg goes green here automatically.

[ronaldtse]

Approving — the fix is correct and validated end to end (regression test fails on main, passes with the change; no other open PR touches this function).

[ronaldtse]

All workflows here should pass (except fuzzing jobs due to google/oss-fuzz#15882 ). Will merge when it is done.

[ronaldtse]

Thank you @MakerYuichi for contributing the fix!