RFR JDK-8131023: JShell: System.in does not work

[Robert Field]

Thumbs up!

-Robert

On 08/28/16 13:26, Jan Lahoda wrote:
> Hi,
>
> Langtools patch updated to reflect the comments is here:
> http://cr.openjdk.java.net/~jlahoda/8131023/langtools.01/
>
> How does it look?
>
> Thanks,
>     Jan
>
> On 26.8.2016 20:11, Robert Field wrote:
>> Continued...
>>
>> MultiplexingOutputStream.java
>>
>> Was this changed so that it occurs in a single write that won't get
>> interleaved?  Do we have this issue elsewhere?  The encoding is now
>> complex enough that the special unrolling for 'void write(int b)' isn't
>> justified.  Could be just:
>>
>> byte[] a = { b };
>> write(a, 0, 1);
>>
>> Thumbs up if those two issues (above and below) are addressed.
>>
>> -Robert
>>
>> On 08/25/16 10:37, Robert Field wrote:
>>> As the bug title states, System.in does not work from Snippets.  In a
>>> command line tool there is only one "in", even though the constructor
>>> for the tool takes a separate cmdIn and userIn.  So, for a command
>>> line tool to work these need to be unified. However, though the JShell
>>> API is the recommended approach to creating new tools, some will
>>> choose to wrap the tool, say in a GUI.  In fact this has already been
>>> done.  Such a wrapped tool can have separate input channels. Rather
>>> than changing the constructor for JShellTool to remove the userIn
>>> parameter, it would seem to leave more options open if a new
>>> constructor with just one input parameter (not named "cmdIn") was
>>> added which forwards to the old constructor.
>>>
>>> More later...
>>>
>>>
>>> On 08/25/16 08:23, Jan Lahoda wrote:
>>>> Hi,
>>>>
>>>> Bug:
>>>> https://bugs.openjdk.java.net/browse/JDK-8131023
>>>>
>>>> Langtools webrev:
>>>> http://cr.openjdk.java.net/~jlahoda/8131023/langtools.00/
>>>>
>>>> JDK repository webrev:
>>>> http://cr.openjdk.java.net/~jlahoda/8131023/jdk.00/
>>>>
>>>> Specdiff:
>>>> http://cr.openjdk.java.net/~jlahoda/8131023/diff.00/jdk/jshell/execution/Util.html 
>>>>
>>>>
>>>>
>>>> Currently jshell forwards output from the agent to the main process.
>>>> With this patch, the input is also forwarded from the main process to
>>>> the agent. InputStream.read() is only called in the main process if
>>>> it has been called on the agent side; this is achieved by using an
>>>> artificial OutputStream in the agent->main process direction. On the
>>>> agent side, when an input is requested, a character is written to the
>>>> corresponding artificial OutputStream, the main process is about
>>>> that, reads from the actual input and sends the data back to the 
>>>> agent.
>>>>
>>>> In the tool, jline is used to actually read the input, as System.in
>>>> is switched in the raw mode. This requires a tweak in jline, to
>>>> detect the proper column on which the editing starts, to cover
>>>> requests like:
>>>>
>>>> System.err.print("Prompt: "); System.err.flush(); System.in.read();
>>>>
>>>> Without the tweak, the line editor wouldn't unfortunately know about
>>>> the text written to the console (for supported terminals, at least).
>>>> (This tweak may be useful in any case for complex prompts that use
>>>> escape sequences; also for jjs.)
>>>>
>>>> An alternative would be to try to switch jline off while reading the
>>>> user's input. The advantage of using jline is that it allows to
>>>> detect Ctrl-C.
>>>>
>>>> Any feedback is welcome!
>>>>
>>>> Thanks,
>>>>      Jan
>>>
>>