Re: [PATCH] Implement efficient 'scm_unget_bytes' and 'unget-bytevector'

[Mark H Weaver]

Mike Gran <address@hidden> writes:

> It is true that none of Guile's clients ever look at putback_buf?

I'm assuming the worst: that Guile's clients might look at, and
manipulate the 'putback_buf' directly.  The way I'm filling
'putback_buf' can *already* happen today.

Here's how it can happen today: unget enough bytes to exactly fill
'putback_buf', and then read some of those bytes.  That will result in
*exactly* the same putback buffer state as what I'm doing here.

> Have you checked that it works with gnome-vfs-port in guile-gnome?

*sigh*  I just looked at that code, and it doesn't do anything the
least bit unusual with the buffers.  No, I didn't run a test with that
code, but even if I did, that wouldn't answer the question of whether
there exists code out there that would be broken by this change.  For
that, only the logic above can help.

Again, the state I'm putting the buffers in can already happen today.
Any code that can't cope with that state is already a ticking time bomb.

      Mark