eshell and after-change-functions

eshell and after-change-functions

[Jan Moringen]

Hi,

while working on a new feature for Rudel (an environment for realtime
collaborative editing in Emacs, [1]), I encountered a problem that has
its root cause in some eshell code, which I can't understand. 

The feature in question is sharing interactive buffers like comint,
slime and, well, eshell with other users. The problem is that Rudel
relies on `after-change-functions' to monitor buffer changes and
propagate them to peers, but eshell let-binds this variable to nil in
`eshell-send-input' and `eshell-output-filter'. This prevents buffer
changes performed in these functions from being propagated and
de-synchronizes the session. I cannot find a specific reason for
disabling all after-change functions in the ehsell code and was
wondering if somebody knows why this is necessary. If it is not
necessary, could this maybe be changed?

Thanks in advance.

Kind regards,
Jan

[1] http://rudel.sourceforge.net

Re: eshell and after-change-functions

[Stefan Monnier]

> The feature in question is sharing interactive buffers like comint,
> slime and, well, eshell with other users. The problem is that Rudel
> relies on `after-change-functions' to monitor buffer changes and
> propagate them to peers, but eshell let-binds this variable to nil in
> `eshell-send-input' and `eshell-output-filter'. This prevents buffer
> changes performed in these functions from being propagated and
> de-synchronizes the session. I cannot find a specific reason for
> disabling all after-change functions in the ehsell code and was
> wondering if somebody knows why this is necessary. If it is not
> necessary, could this maybe be changed?

Have you tried to remove the offending let-binding and see what effect
it has on Eshell's behavior?
The commit logs don't give any help about the reason for these bindings,
and looking at the code I can't figure it out myself either, so unless
John remembers why they were added, we'll be forced to just "try and
pray".


        Stefan

Re: eshell and after-change-functions

[Jan Moringen]

Hi Stefan.

> > The feature in question is sharing interactive buffers like comint,
> > slime and, well, eshell with other users. The problem is that Rudel
> > relies on `after-change-functions' to monitor buffer changes and
> > propagate them to peers, but eshell let-binds this variable to nil in
> > `eshell-send-input' and `eshell-output-filter'. This prevents buffer
> > changes performed in these functions from being propagated and
> > de-synchronizes the session. I cannot find a specific reason for
> > disabling all after-change functions in the ehsell code and was
> > wondering if somebody knows why this is necessary. If it is not
> > necessary, could this maybe be changed?
> 
> Have you tried to remove the offending let-binding and see what effect
> it has on Eshell's behavior?

Removing the let-binding solves the problem for Rudel. A brief test did
not reveal any problems with eshell either, but a more experienced
developer should be the judge of that.

> The commit logs don't give any help about the reason for these bindings,
> and looking at the code I can't figure it out myself either, so unless
> John remembers why they were added, we'll be forced to just "try and
> pray".

Thanks. This would be very hard or even impossible to work around by
sensible means.

Jan