Profiles filtering by ahresse · Pull Request #378 · fluendo/fluster

ahresse · 2026-07-09T00:00:31Z

This pull request adds support for filtering test vectors by codec profile, allowing users to specify which profiles to run via a new -p/--profiles command-line option. The feature is fully documented, supports multiple profiles, is case-insensitive, and includes automatic inference of VP9 profiles from test vector names. The implementation also introduces new profile values for VP9 in the Profile enum and ensures that only matching test vectors are executed when filtering is active.

Profile Filtering Feature:

Added a -p/--profiles argument to the CLI, enabling users to run only test vectors matching specified codec profiles; the option supports multiple, case-insensitive profile names and is documented in README.md [1] [2] [3] [4].
Updated context and initialization logic in fluster/fluster.py to parse, validate, and store user-supplied profile names, mapping them to the new or existing Profile enum values; errors are reported for unknown profiles [1] [2] [3] [4] [5].

VP9 Profile Support:

Added VP9 profile values (VP9 Profile 0 through VP9 Profile 3) to the Profile enum in fluster/codec.py.
Implemented automatic VP9 profile inference from test vector names (e.g., vp90- prefix → Profile 0) in TestSuite, and assigned inferred profiles to test vectors when loading VP9 test suites from JSON [1] [2].

Test Vector Filtering Logic:

Updated TestSuite.generate_tests to filter out test vectors whose profiles do not match the requested profiles, ensuring only relevant vectors are run.

Interface and Internal Consistency:

Propagated profile filtering information through context objects and constructors in fluster/fluster.py and fluster/test_suite.py, ensuring consistent handling across the codebase [1] [2] [3].

ylatuya

Thanks for your first contribution! Filtering by profiles is super helpful.

Do you think this filter should also be enabled for the list command? It will also help list test vectors that match a specific profile.

ahresse · 2026-07-10T00:54:12Z

@ylatuya I implemented your proposition in last commit. Quite useful to be able to list profiles indeed.

mdimopoulos · 2026-07-16T10:49:14Z

+    CONSTRAINED_BASELINE = ("Constrained Baseline", Codec.H264)
+    BASELINE = ("Baseline", Codec.H264)
+    EXTENDED = ("Extended", Codec.H264)
+    MAIN = ("Main", Codec.H264)
+    HIGH = ("High", Codec.H264)
+    HIGH_10 = ("High 10", Codec.H264)
+    HIGH_10_INTRA = ("High 10 Intra", Codec.H264)
+    HIGH_4_2_2 = ("High 4:2:2", Codec.H264)
+    HIGH_4_2_2_INTRA = ("High 4:2:2 Intra", Codec.H264)
+    HIGH_4_4_4_INTRA = ("High 4:4:4 Intra", Codec.H264)
+    HIGH_4_4_4_PREDICTIVE = ("High 4:4:4 Predictive", Codec.H264)
+    CAVLC_4_4_4 = ("CAVLC 4:4:4", Codec.H264)
+    CAVLC_4_4_4_INTRA = ("CAVLC 4:4:4 Intra", Codec.H264)
+
+    MAIN_10 = ("Main 10", Codec.H264)
+    MAIN_STILL_PICTURE = ("Main Still Picture", Codec.H264)
+    MAIN_4_2_2_10 = ("Main 4:2:2 10", Codec.H264)
+    MAIN_4_4_4_12 = ("Main 4:4:4 12", Codec.H264)


Some of these are definitely shared with H265 (and possibly H266).
Sorry, the comment above this piece of code is obsolete. I can't make a proper evaluation of the impact of the PR right now but please take it into account

cc @ylatuya

Right, profiles are shared among different codecs and they should be associated to a list of codecs instead of single codec.

ylatuya · 2026-07-16T11:01:47Z

+    CONSTRAINED_BASELINE = ("Constrained Baseline", Codec.H264)
+    BASELINE = ("Baseline", Codec.H264)
+    EXTENDED = ("Extended", Codec.H264)
+    MAIN = ("Main", Codec.H264)
+    HIGH = ("High", Codec.H264)
+    HIGH_10 = ("High 10", Codec.H264)
+    HIGH_10_INTRA = ("High 10 Intra", Codec.H264)
+    HIGH_4_2_2 = ("High 4:2:2", Codec.H264)
+    HIGH_4_2_2_INTRA = ("High 4:2:2 Intra", Codec.H264)
+    HIGH_4_4_4_INTRA = ("High 4:4:4 Intra", Codec.H264)
+    HIGH_4_4_4_PREDICTIVE = ("High 4:4:4 Predictive", Codec.H264)
+    CAVLC_4_4_4 = ("CAVLC 4:4:4", Codec.H264)
+    CAVLC_4_4_4_INTRA = ("CAVLC 4:4:4 Intra", Codec.H264)
+
+    MAIN_10 = ("Main 10", Codec.H264)
+    MAIN_STILL_PICTURE = ("Main Still Picture", Codec.H264)
+    MAIN_4_2_2_10 = ("Main 4:2:2 10", Codec.H264)
+    MAIN_4_4_4_12 = ("Main 4:4:4 12", Codec.H264)


Right, profiles are shared among different codecs and they should be associated to a list of codecs instead of single codec.

ahresse · 2026-08-17T10:30:25Z

@ylatuya and @mdimopoulos: Last commit might fulfil your previous comments.

profile 0: color depth: 8 bit/sample, chroma subsampling: 4:2:0 profile 1: color depth: 8 bit, chroma subsampling: 4:2:2, 4:2:0, 4:4:4 profile 2: color depth: 10–12 bit, chroma subsampling: 4:2:0 profile 3: color depth: 10–12 bit, chroma subsampling: 4:2:2, 4:2:0, 4:4:4 Signed-off-by: Alexandre Esse <alexandre.esse@canonical.com>

to filter test vectors by codec profile

Signed-off-by: Alexandre Esse <alexandre.esse@canonical.com>

- Add codec attribute to Profile enum - Profile members now carry a codec attribute via tuple values with a custom __new__ - list command shows profiles grouped by codec (filterable with -d/--codec) - TestVector stores/serializes codec alongside profile Signed-off-by: Alexandre Esse <alexandre.esse@canonical.com>

Redefine each Profile member to carry a list of codecs instead of a single one, so profiles common to several codec families can be shared, e.g. "Main" (H.264, H.265, AV1, MPEG2) and "Main 10" (H.265, H.266). Signed-off-by: Alexandre Esse <alexandre.esse@canonical.com>

The codec field was only serialized next to the profile and never read back as a Codec enum, causing a type mismatch on reload. Remove it entirely since the test suite already carries the codec. Signed-off-by: Alexandre Esse <alexandre.esse@canonical.com>

AV1 defines Main, High and Professional profiles. Main already maps to Codec.AV1, so add AV1 to the shared High profile for a complete profile listing. Signed-off-by: Alexandre Esse <alexandre.esse@canonical.com>

Add the -p/--profiles option to the reference subcommand so reference runs can be restricted to specific codec profiles, mirroring run. Signed-off-by: Alexandre Esse <alexandre.esse@canonical.com>

Signed-off-by: Alexandre Esse <alexandre.esse@canonical.com>

Match the test_vector_names pattern by storing the profiles to filter as a Set instead of a List. Signed-off-by: Alexandre Esse <alexandre.esse@canonical.com>

Sort codecs and their profiles alphabetically in list_profiles for a deterministic listing. Signed-off-by: Alexandre Esse <alexandre.esse@canonical.com>

ahresse · 2026-08-17T17:37:38Z

I suspect CI to fail because of network failures. (Github is experiencing degraded performance right now. (https://www.githubstatus.com/) - make check "works on my machine" :-)

ylatuya reviewed Jul 9, 2026

View reviewed changes

Comment thread fluster/codec.py Outdated

ylatuya mentioned this pull request Jul 9, 2026

Need advice on creating a vp9 suite specific to Apple VideoToolbox in GStreamer #329

Open

ahresse force-pushed the profiles-filtering branch from 15538ff to 75782a9 Compare July 9, 2026 14:27

ahresse force-pushed the profiles-filtering branch 2 times, most recently from 27c11b9 to f8f6c0f Compare July 10, 2026 09:44

ylatuya approved these changes Jul 16, 2026

View reviewed changes

mdimopoulos reviewed Jul 16, 2026

View reviewed changes

ylatuya requested changes Jul 16, 2026

View reviewed changes

ahresse force-pushed the profiles-filtering branch from f8f6c0f to 339ae09 Compare August 4, 2026 10:47

ahresse added 5 commits August 17, 2026 15:05

feat: add run --profiles CLI option

b3dfd02

to filter test vectors by codec profile

docs: add profile options instructions to README

eaf71b2

Signed-off-by: Alexandre Esse <alexandre.esse@canonical.com>

ahresse force-pushed the profiles-filtering branch from dbe9c53 to 83a80f1 Compare August 17, 2026 13:05

ahresse added 6 commits August 17, 2026 16:50

fix: add AV1 to the High profile

30a6bd0

AV1 defines Main, High and Professional profiles. Main already maps to Codec.AV1, so add AV1 to the shared High profile for a complete profile listing. Signed-off-by: Alexandre Esse <alexandre.esse@canonical.com>

feat: support profile filtering in the reference subcommand

7b4fcea

Add the -p/--profiles option to the reference subcommand so reference runs can be restricted to specific codec profiles, mirroring run. Signed-off-by: Alexandre Esse <alexandre.esse@canonical.com>

chore: ignore .venv directories

d770eba

Signed-off-by: Alexandre Esse <alexandre.esse@canonical.com>

refactor: use a Set for the profile filter

8a3d059

Match the test_vector_names pattern by storing the profiles to filter as a Set instead of a List. Signed-off-by: Alexandre Esse <alexandre.esse@canonical.com>

chore: sort profiles in the list output

1e94f29

Sort codecs and their profiles alphabetically in list_profiles for a deterministic listing. Signed-off-by: Alexandre Esse <alexandre.esse@canonical.com>

Navigation Menu

Uh oh!

Profiles filtering - #378

Profiles filtering#378
ahresse wants to merge 11 commits into
fluendo:masterfrom
ahresse:profiles-filtering

ahresse commented Jul 9, 2026

Uh oh!

ylatuya left a comment •

edited

Loading

Uh oh!

Uh oh!

ahresse commented Jul 10, 2026

Uh oh!

mdimopoulos Jul 16, 2026 •

edited

Loading

Uh oh!

ylatuya Jul 16, 2026

Uh oh!

ylatuya Jul 16, 2026

Uh oh!

ahresse commented Aug 17, 2026

Uh oh!

ahresse commented Aug 17, 2026 •

edited

Loading

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

3 participants

Uh oh!

Conversation

ahresse commented Jul 9, 2026

Uh oh!

ylatuya left a comment • 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!

Uh oh!

ahresse commented Jul 10, 2026

Uh oh!

mdimopoulos Jul 16, 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!

ylatuya Jul 16, 2026

Choose a reason for hiding this comment

Uh oh!

ylatuya Jul 16, 2026

Choose a reason for hiding this comment

Uh oh!

ahresse commented Aug 17, 2026

Uh oh!

ahresse commented Aug 17, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

3 participants

ylatuya left a comment •

edited

Loading

mdimopoulos Jul 16, 2026 •

edited

Loading

ahresse commented Aug 17, 2026 •

edited

Loading