If I am reading the regex correctly, it looks like the whole src:value part is totally removed. I don't think this is a good idea. I think code may want to know the src, and the fact that it contained base64. Really, we just want the bit after the comma (,) to be truncated. Apologies if I'm reading it incorrectly.

To be clear, this function never receives "text content". Rather, it only receives well-formatted IAccessible2 attribute strings. I think the confusion here is that we get attribute strings both from IAccessible2 objects, and from format runs of IAccessible2 text. But in both cases, we are talking about name:value;name:value; strings, not random text content.

Still, this pr does make the regex more specific, only applying to the src attribute, which is a good thing.

Thanks for this clarification.
I was conscious of the "name:value;name:value" aspect of the passed string, but was afraid one of these values might be actual text content, in which case we could have had a nasty collision if the user was indeed using an online editor to edit a web page.

If I am reading the regex correctly, it looks like the whole src:value part is totally removed. I don't think this is a good idea. I think code may want to know the src, and the fact that it contained base64. Really, we just want the bit after the comma (,) to be truncated. Apologies if I'm reading it incorrectly.

No problem, regex are always a challenge to read, and this one especially requires braille or total speech verbosity to be perceived correctly.
The first part of the replacement string is actually a reference to a capturing group.
Thus, the resulting string is of the form "data:image/jpg;base64<truncated>", that is, it preserves the "data" scheme and the original MIME type of the base64 data.
Nevertheless, I realize that it strips out the comma between "base64" and "<truncated>", but I can alter it so that this comma is preserved too.

On the contrary, my vbuf C++ proposals do not produce that same result as they also strip out the MIME type and produce results of the form "data:<truncated>". If you feel this is problematic, I'll update the proposed implementation so that the original MIME type is preserved.

EDIT: Escaped the angle brackets for clarity.