Fix init/state registration correctness

[mulugetam]

• deflateInit_, deflateInit2_, inflateInit_, inflateInit2_: call the original zlib init function first and only register stream settings on Z_OK. Previously settings were registered unconditionally, so a failed init left a stale entry keyed by a z_streamp the app may free (it never calls the matching End), and a rejected re-init overwrote the settings of a stream zlib had left valid and usable.

• gzeof: two fixes, same root cause — reached_eof was trusted where the shim does not fully maintain it.

  • Fall through to orig_gzeof() when the file has no map entry or is on the ZLIB path. reached_eof is only maintained by the accelerator read path, so on the zlib path it stayed false forever and gzeof never reported EOF, causing infinite read loops.
  • On the accelerator path, do not report EOF while data_buf/io_buf still hold bytes gzread has not returned. reached_eof only records that a read() of the file came up short, so while (!gzeof(fp)) gzread(...) exited early and silently truncated with a success return. Measured: 4096 of 8192 bytes on an 8KB file read in 4KB chunks. gzeof now checks the same buffer state gzread already does.

Tests: two GzipFileTest cases, one per gzeof fix, each verified to fail with its hunk reverted. Full suite with QAT+IAA+IGZIP on real hardware: 28703 tests, 24383 passed, 4320 skipped, 0 failed.

Rebased onto main after #65; depends on its null guards, since conditional registration means a failed *Init leaves no map entry.

[Copilot]

Pull request overview

This PR fixes correctness issues in the zlib interposition layer by ensuring per-stream/per-file shim state is only registered/used when zlib has actually accepted/owns that state, preventing stale map entries and incorrect EOF reporting.

Changes:

• Make deflateInit_ / deflateInit2_ / inflateInit_ / inflateInit2_ register stream settings only when the underlying zlib init returns Z_OK.
• Fix gzeof to defer to orig_gzeof() on the zlib path (or when unregistered), and to avoid reporting EOF while shim buffers still contain unread data.
• Add unit tests covering both gzeof fixes and update an existing test comment to reflect the new init/registration behavior.

Reviewed changes

Copilot reviewed 2 out of 2 changed files in this pull request and generated 1 comment.

File	Description
zlib_accel.cpp	Registers per-stream settings only on successful init; fixes gzeof behavior for both zlib and accelerator paths.
tests/zlib_accel_test.cpp	Adds GzipFileTest coverage for gzeof on zlib path and buffered-data cases; updates existing test rationale text.

Suppressed comments (1)

tests/zlib_accel_test.cpp:1531

• Same issue here: the test updates global config and relies on reaching the bottom of the function to restore it. A failing ASSERT in the middle will leave non-default config set for subsequent tests. Add an on-scope-exit restorer right after saving the values.

  const uint32_t saved_iaa = GetConfig(USE_IAA_UNCOMPRESS);
  const uint32_t saved_qat = GetConfig(USE_QAT_UNCOMPRESS);
  const uint32_t saved_zlib = GetConfig(USE_ZLIB_UNCOMPRESS);
  const uint32_t saved_iaa_compress = GetConfig(USE_IAA_COMPRESS);
  const uint32_t saved_qat_compress = GetConfig(USE_QAT_COMPRESS);

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

[asonje]

lgtm