MCAP cli produces empty message index

Description
The MCAP cli produces an empty message index instead of omitting it. It seems that this slows down Foxglove and the Python lib.

Version: CLI v0.0.24
Platform: mac OS

Steps To Reproduce
mcap compress on any file

Expected Behavior
no message index at all for topics not in this chunk

Internal tracking ticket: FG-2527

Hi, thanks for the bug report, however, the "steps to reproduce" section is lacking. Can you provide more information about how one can see that this issue has occurred? And can you provide more detail about what slowdown you have observed?

Sorry for the unclear explanation, for example the slow down occures in the following code:

with open(sys.argv[1], "rb") as f:
    reader = make_reader(f)
    for schema, channel, message in reader.iter_messages(topics=["/tf_static"]):
        print(f"{channel.topic} ({schema.name}): {message.data}")

even after printing all /tf_static messages the reader keeps reading the whole file.
This only happens when the file was previously processed by the mcap cli (compress, filter, recover, ...), it does not happen for example when the mcap was produced by ros2 bag record
I also noticed slowdowns in foxglove studio when skipping in large files.

I've checked the record Length of the MessageIndex here

mcap/python/mcap/mcap/reader.py

Lines 84 to 86 in 0842cbf

    
           if topics is None or summary.channels[channel_id].topic in topics: 
        
               out.append(chunk_index) 
        
               break

which showed that a MessageIndex for /tf_static exists for all chunks but the record length is almost always zero.

The code snippet is helpful, can you share an example file as well?

	if topics is None or summary.channels[channel_id].topic in topics:
	out.append(chunk_index)
	break