Dependency Refactoring: element retriction in api.git

[myungjoo]

Element restriction of Inference.Pipeline APIs should be done independently in api.git, not in nnstreamer.git

This addresses nnstreamer/nnstreamer#3553

The next commit should address

1. unit testing: unittest_capi_inference.cc nnstreamer_capi_util element_available_*

2. packaging (install the .txt file along with api rpms)

3. removing restriction mechanisms from nnstreamer.git

[taos-ci]

📝 TAOS-CI Version: 1.5.20200925. Thank you for submitting PR #409. Please a submit 1commit/1PR (one commit per one PR) policy to get comments quickly from reviewers. Your PR must pass all verificiation processes of cibot before starting a review process from reviewers. If you are new member to join this project, please read manuals in documentation folder and wiki page. In order to monitor a progress status of your PR in more detail, visit http://ci.nnstreamer.ai/.

[taos-ci]

cibot: @myungjoo, c/src/ml-api-inference-pipeline.c includes bug(s). Please fix incorrect coding constructs in your commit before entering a review process.

[taos-ci]

cibot: @myungjoo, A builder checker could not be completed because one of the checkers is not completed. In order to find out a reason, please go to http://ci.nnstreamer.ai/nnstreamer-api/ci/repo-workers/pr-checker/409-202310172342550.85050296783447-058a77f954c47365688e5842573c7278af178493/.

[myungjoo-bot]

Automated review of this draft (transcribed from an AI review agent's report; please verify before acting).

Summary: The commit replaces the nnsconf_get_custom_value_*("element-restriction", ...) lookup in _ml_check_plugin_availability() with a hard-coded allow-list file /etc/ml_inference_pipeline_allowed_elements.txt parsed by new static helpers, so api.git no longer depends on nnstreamer's conf module (nnstreamer/nnstreamer#3553). The direction is still valid and not superseded: main's _ml_check_plugin_availability (c/src/ml-api-inference-pipeline.c:784-833) is identical to the merge-base version, git merge-tree is conflict-free, api.git still has no file-based mechanism, nnstreamer.git still carries the whole mechanism (enable-element-restriction, spec-generated nnstreamer.ini), and issue 3553 is still open. However, this 2023 sketch does not compile, and if it did the allow-check would reject every non-tensor_* element. It needs a rewrite with tests and packaging in the same PR or a stacked series.

All file:line references are PR-side c/src/ml-api-inference-pipeline.c.

1. [High] Does not compile — :844 if (valid_lines_check => valid_lines): => is not a C operator. Fix: >=.
2. [High] Stray semicolons make the allow-check always fail — :878 if (is_element_restriction_valid == FALSE); and :881 if (is_element_restricted == FALSE); are empty statements, so return FALSE; at :879 runs unconditionally. Every element except nnstreamer / tensor_* is reported restricted, ml_pipeline_construct returns ML_ERROR_NOT_SUPPORTED for essentially every test pipeline, and ml_check_element_availability reports everything unavailable. Fix: drop the semicolons; add a test constructing videotestsrc ! tensor_converter ! fakesink with no allow-list present.
3. [High] printf argument mismatch — :822: two %s but only err->message is passed (filepath missing); reads a garbage pointer at runtime. Fix: pass filepath too.
4. [High] Uninitialized err / contents on the read-failure path — :800, :818-825: GError *err; is never NULLed, and if g_file_get_contents fails with err == NULL control falls through to g_strsplit (contents, ...) at :828 with contents uninitialized. Fix: initialize both to NULL and return unconditionally on failure after g_clear_error.
5. [High] Fail-open default is a Tizen regression given the missing packaging — :811-815: file absent => allow everything. No package in api.git installs the .txt (packaging/machine-learning-api.spec has no entry), and nnstreamer's [element-restriction] ini section, still generated on Tizen, is no longer read by anyone after this diff. Net effect on Tizen: restriction silently turns off. Fix: land packaging in the same PR (list generated from the spec %define allowed_element_* macros, root-owned 0644), and on __TIZEN__ treat a missing file as deny (or at least _ml_loge), keeping fail-open only for non-Tizen dev builds.
6. [Medium] Not thread-safe; reload storm on unreadable file — :873-876: allowed_elements and the two flag statics are mutated without a lock; two threads on first use can g_strfreev an array the other is iterating (UAF). When the file exists but is unreadable, is_element_restriction_valid stays FALSE so the file is re-opened for every element of every construct. Fix: load once under g_once_init_enter/leave (or a GMutex), and make the unreadable case terminal (deny all, valid=TRUE).
7. [Medium] Brittle parsing — :828-857: g_strsplit (contents, "\n") leaves \r on CRLF files, trailing whitespace is kept, whitespace-only lines count as elements, # is only honored at column 0, and find_key_strv needs an exact case-folded match so "appsrc\r" never matches. Fix: g_strstrip each line, skip empty / #-prefixed lines after stripping, accumulate into a GPtrArray; this also removes the two-pass count and the "changed at run-time" branch.
8. [Medium] Existing tests check a different config source than the library — tests/capi/unittest_capi_inference.cc:2382 still gates element_available_01_p on nnsconf_get_custom_value_bool ("element-restriction", ...), which no longer relates to library behavior. Fix: tests that write a temp allow-list via an internal test-only setter (not an env var), and assert allowed/denied, empty file => deny all but tensor_*, comments/CRLF tolerated, file absent => allow all on non-Tizen.
9. [Low] Identifiers and style — __load_allowed_elements / __is_element_allowed (:795, :873): __-prefixed identifiers are reserved in C and the file uses _ml_*; return type on the signature line and gchar *filepath violate gst-indent (what cibot flagged). Typos: "alloed" (:810), "laod" (:845), "retriction" (title).
10. [Low] No documentation of the file contract — path, one-element-per-line format, # comments, absent => allow all, empty => deny all. Add a short section in Documentation/ and a @remarks on ml_check_element_availability.
11. [Low] Dependency goal only partially met — the function still relies on find_key_strv from nnstreamer_plugin_api.h, and unittest_capi_inference_single.cc still calls nnsconf_get_custom_value_string in seven places. A local g_strv_contains-style loop would drop find_key_strv too.

No back-door found. The new code consults no environment variable or user-writable path; the only input is a root-owned /etc file. Compared with main it actually removes the NNSTREAMER_CONF env-influenced lookup, which is good; please keep it that way and use an internal test-only setter rather than an env override for tests.