With GGML_BACKEND_DL=ON no backend is registered until something calls
ggml_backend_load_all(). Every example does; the tests never did, so
whisper_init_* and parakeet_init_* ran with devices = 0 and aborted on
GGML_ASSERT(device) in ggml_backend_dev_backend_reg.
Three tests aborted on such a build - test-whisper-zero-samples,
test-vad and test-parakeet - taking ctest -L gh to 2 of 4.
test-vad-full has the same fault and reaches it only once a real
base.en model is present.
No workflow caught this because none pairs the two: build-clang,
build-gcc and build-sanitize run ctest -L gh but never set
GGML_BACKEND_DL, while release.yml sets it and runs no tests.
ggml_backend_load_all() is already reachable through whisper.h and
parakeet.h via ggml-cpu.h, so no new include is needed.
Resolves: https://github.com/ggml-org/whisper.cpp/issues/4030
whisper_full()/whisper_full_with_state() only compute the mel spectrogram when
n_samples > 0. For n_samples == 0 on a freshly allocated state the mel is never
touched, but struct whisper_mel had no member initializers, so n_len / n_len_org
/ n_mel were indeterminate heap garbage (the state is allocated with
new whisper_state). seek_end is derived from that garbage and, depending on it,
the call either quietly returns 0 or runs the encoder with garbage dimensions
over an empty (NULL) mel buffer, dereferencing address 0 in the mel copy loop.
Give whisper_mel default member initializers so a never-computed mel reads as
0 frames and the n_samples == 0 case deterministically takes the existing
too-short path.
Fixes#3978
Abir Deol