Re: sciTe editor IRB window getting double characters

David Boyd:

> a) Clicking into the output pane (which gains focus then).
> Now you can type in some string, enter return two times and all 
> output is shown then.

    The way that SciTE's output pane works is that it has two file 
handles open: one writing to the subprocess and one reading from it. 
Characters typed are added to the buffer and then written to the write 
pipe.

    Data from the read pipe is added to the display when it is received. 
Programs often buffer output differently when writing to a pipe (maximum 
buffering) and writing to a console (line buffering) as a console 
implies a human interacting with the program who needs to see prompts. 
SciTE has no way to stop the subprocess buffering data (there is no 
equivalent of a pseudo-terminal on Windows) except for invocation 
options. Some programs such as Python have flags (-u) to turn off 
buffering. For other languages, there is often a call like C's setvbuf 
that can be used to change buffering on a stream.

> You have to insert '$stdout.flush' here too after each 'puts' line.
> Also the cursor stays in the code pane and focus must be changed by
> explicitly clicking into the output pane.

    SciTE can not see that the subprocess wants to read input. There 
could be an option to switch focus when running particular applications 
but for many commands like build it is better for the focus to stay in 
the editor pane. A focus switching option also needs to handle whether 
to return focus to the edit pane after command completion.

> SOLUTION: is to simply change the subsystem to '2'. An extra Window is 
> opened and all works as expected.

    Subsystem 2 is ShellExecute which has its own set of idiosyncrasies.

> Starting irb with the command setup as in ruby.properties opens the irb prompt
> in the scite output pane, but every input is echoed double.

    This means that irb is echoing input. The ordering of the events in 
SciTE make it difficult to change from the SciTE end. SciTE is 
performing the character write upon receiving the SCN_CHARADDED 
notification where the character has already been added to the buffer. 
The reason it uses SCN_CHARADDED is because it is sent for cooked 
characters after processing commands.

    It may be possible to switch to reading key events and sending 
through to the subprocess while running a command but this would be a 
lot of work.

    The biggest problem with SciTE's output pane for many is that it 
interprets command characters. For example, typing ab[Backspace]c sends 
the three characters abc through to the subprocess although the pane 
shows ac. It has been proposed that there could be a line buffering 
option where SciTE would build up a whole line and only send it when 
Enter is pressed to handle this.

    SciTE's output pane is not clever: the original goal in making it 
interactive was to be able to respond to CVS prompts. Possible changes 
will often make other scenarios fail so should be implemented as 
options. As the output pane is currently sufficient for my needs, I 
won't be working on improvements.

    Neil