[TCLCORE] Dropping into standard repl from script - piperepl patch

[<jul...@pr...>]

Hi All,
There are occasionally requests for how to drop into the tclsh repl from 
a script - perhaps after an error for debugging.
The usual advice seems to be to create a small repl in script. This is 
at least inconvenient and probably not compatible with uparrow history 
etc(?)

The first inclination probably is to play with ::tcl_interactive to see 
if that will help - but it doesn't.
In Tcl_MainEx - ::tcl_interactive is mapped via Tcl_Linkvar to the 
is.tty c value  - but unfortunately (namewise at least) this isn't a 
direct  correspondence to what people would consider isatty.
This is because in Tcl it is set to zero when a script is run (as 
documented)  - even though (when not part of a pipeline) there can still 
be a terminal.

As I also have a usecase for dropping into the *real* repl - I've 
developed a patch for tclMain.c to experiment with this. (developed 
against latest tcl9)

I'm referring to this patch project as: piperepl
It is only active if ::env(TCLSH_PIPEREPL) is set true. (with minor 
effects re tclsh array noted below)

It uses some linked variables in  a global ::tclsh array to allow a 
script to very simply set whether the repl should run after the script. 
(and possibly whether stdin should be evaluated as commands)

In the standard 'tclsh <script>' case It does this by dropping through 
to the normal repl code, and in the case of a pipeline, by reopening 
stdin on the console after we get eof on stdin.
For example error.tcl  - the first 2 lines enable the behaviour.

    ------- error.tcl -------
    set ::tclsh(dorepl) 1
    set ::tcl_interactive 1 ;#Not required - but not as pretty without it
    set myval whatever
    error "something is wrong"
    --------------------------

With ::env(TCLSH_PIPEREPL) = 1 you can now drop straight into the repl 
and examine variables errors etc.

If you use:
printf "puts test" | tclsh
Like standard tclsh it will just evaluate "puts test"

printf "set ::tclsh(dorepl) 1; puts test" | tclsh
will evaluate the script from stdin and also drop into the repl - but 
unless you also set ::tcl_interactive 1 in the printf statement - you'll 
get it without prompts.

For security reasons.. (and consistency with standard tclsh)
printf "puts test" | tclsh <script>
will not evaluate the printf script from stdin - and in this case you 
can't of course influence it with ::tclsh(dorepl) in the piped in value
This makes sense because the data being piped in to a script shouldn't 
be assumed to be commands for evaluation anyway
The data on stdin may or may not be read by the script - but any 
remaining unconsumed is currently just stored in ::tclsh(inputbuffer)
A script can decide it does want the stdin from the pipeline to be 
evaled  by setting ::tclsh(evalinput)  (but there is deliberately no way 
to set that from the stdin pipeline in this case)

In the above error.tcl case - it may be inconvienent to have 
tclsh(dorepl) true and always drop into the shell
By setting tclsh(evalinput) 1 in the .tcl script instead of 
tclsh(dorepl) - you can run tclsh error.tcl  in the normal terminating 
manner - and call it as printf "set tclsh(dorepl) 1; set tcl_interactive 
1" | tclsh error.tcl  when you want the full repl.


I believe the result is that even with the TCLSH_PIPEREPL environment 
variable enabled - when the tclsh(x) vars aren't set in scripts it 
should be consistent with standard tclsh - aside from creating the 
global ::tclsh array
Which I understand may be an issue - and am open to renaming/namespacing 
if that is possible - although I think it's quite reasonable for the 
functionality.
The tclsh(istty) variable is I think a useful addition at the script 
level anyway - as tcl_interactive doesn't serve this purpose.

Only tested on windows (but when reopening to console, a line to open 
/dev/tty is included when not on windows - so maybe that works?)

Is anyone interested in this feature - and does anyone here have any 
comments regarding feasibility/desirability for something like this to 
be included in future?

I believe, but haven't tested; that it shouldn't have any impact on Tk 
startup or other custom startup but note that I'm not experienced with c 
programming or Tcl internals.
I've read about refcounts and various bits and pieces - and tried to use 
examples from elsewhere in the source to get close to doing the right 
thing - but it definitely needs scrutiny
Code review/comments would be most welcome even if this doesn't look 
like a candidate for inclusion in Tcl proper.

There is a somewhat ugly debugging message in the attached patch to show 
when console reopen is about to happen - left in for evaluation purposes 
show what it's doing.

I've tried to use the same tab/space conventions as the rest of the file 
- but I'm not entirely sure I got that right.
Patch made from cmd.exe with diff -u --binary tclMain.c tclMain_piperepl.c
lineendings are unix lf

Cheers,
Julian (JMN on wiki)