Upon fiddling with conversions, I discovered that a silent test signal just grew enormously (relatively speaking), creating a SEEKTABLE that extends far beyond the file itself. After a bit of debugging it turns out that I had forgotten the --ignore-chunk-sizes from custom settings.
That was (to me) an unexpected effect of that setting. Replicated on 1.6.14 and a couple of 2.0 betas.
* This isn't supposed to happen, no?!
* Even if user is to blame, this isn't outright easy to fix, so ... new feature request: an improved "Fix FLAC seek table"
From the metaflac output, you see the number of samples is < 1.5 million, and seekpoints from number 4 on fall outside that range.
METADATA block #0
type: 0 (STREAMINFO)
is last: false
length: 34
minimum blocksize: 4096 samples
maximum blocksize: 4096 samples
minimum framesize: 14 bytes
maximum framesize: 17 bytes
sample_rate: 44100 Hz
channels: 2
bits-per-sample: 16
total samples: 1499400
MD5 signature: 623850d4a4b305f93262d235167f922f
METADATA block #1
type: 3 (SEEKTABLE)
is last: false
length: 43830
seek points: 2435
point 0: sample_number=0, stream_offset=0, frame_samples=4096
point 1: sample_number=438272, stream_offset=1498, frame_samples=4096
point 2: sample_number=880640, stream_offset=3097, frame_samples=4096
point 3: sample_number=1318912, stream_offset=4702, frame_samples=4096
point 4: sample_number=1764000, stream_offset=0, frame_samples=0
point 5: sample_number=2205000, stream_offset=0, frame_samples=0
point 6: sample_number=2646000, stream_offset=0, frame_samples=0
point 7: sample_number=3087000, stream_offset=0, frame_samples=0
point 8: sample_number=3528000, stream_offset=0, frame_samples=0
point 9: sample_number=3969000, stream_offset=0, frame_samples=0
[etc!]