[pragmatrix]

Why

To record audio reliably, the recording bug has to see each frame after every other bug has changed it. On the write path, SMBF_WRITE_STREAM and SMBF_WRITE_REPLACE callbacks run in one loop in list order, so a recorder only sees the written audio as modified by the bugs linked in front of it. SMBF_FIRST lets a bug go to the front of the list, but nothing lets a bug stay at the end: any bug added later is appended behind it. So once a bug that replaces written frames is added after the recording starts, the recording no longer captures the audio that is actually sent.

The read path doesn't have this problem: all SMBF_READ_REPLACE callbacks run before any SMBF_READ_STREAM buffer is filled. There, SMBF_LAST only orders the replace bugs among themselves.

Summary

• Add SMBF_LAST (1 << 28). A bug added with this flag goes to the end of the list. Bugs added after it, without the flag, are linked in front of the first SMBF_LAST bug, so SMBF_LAST bugs always stay at the tail, in the order they were added.
• switch_core_media_bug_add rejects SMBF_FIRST | SMBF_LAST with SWITCH_STATUS_GENERR.
• The flag is only honored when the bug is added. Setting it later with switch_core_media_bug_set_flag does not move the bug.
• Unit test media_bug_last_ordering in tests/unit/switch_ivr_async.c pins down the ordering together with SMBF_FIRST.

Validation

• Not built or tested locally: the tree isn't configured on the dev machine, and this fork has no CI runs.
• fs-image runs switch_ivr_async in its unit-test stage, so it will cover the new test once it builds from this fork instead of signalwire/freeswitch v1.11.3.

Known gaps

• src/mod/languages/mod_managed/managed/swig.cs has not been regenerated, so SMBF_LAST is missing from the C# bindings.

Note

This was created with AI assistance.