Use simplified keymap for dumb terminal

[hvesalai]

Fixes #864

[hvesalai]

Side note: during fixing this I run into some other problems. Most notably current implementation inhibits CRs from dumb terminal which means if you run

env TERM=dumb ./build repl

The carriage return key will not do anything. The only way to do ACCEPT_LINE is CTRL-J (i.e. LF)

This problem stems from the line https://github.com/jline/jline3/blob/master/reader/src/main/java/org/jline/reader/impl/LineReaderImpl.java#L639

[hvesalai]

Side note 2: the first test in reader/src/test/java/org/jline/terminal/impl/AbstractWindowsTerminalTest.java (testBracketingPasteSmall()) was failing already before I started making this change and I didn't fix it.

It should be noted that it will not work on dumb terminals unless the respective bind is added to the dumb terminal bindings, i.e. bind(dumb, BEGIN_PASTE, BRACKETED_PASTE_BEGIN);

Doing that binding would be nonsensical since bracketed paste is not turned on for dumb
https://github.com/jline/jline3/blob/master/reader/src/main/java/org/jline/reader/impl/LineReaderImpl.java#L631

[gnodet]

Would adding "\r\n" as an additional key binding help with the cr/lf issue on dumb terminals ?

[hvesalai]

That would not be deterministic then. How will the system know if to expect only \r or \r\n. Remember that there might be no immediate input after the \r. Consider the following interaction:

Time n: user inputs something that triggers SELF_INSERTs, such as "Hello"
Time n+1: user inputs \r.
Time n+m: user inputs a new thing that trigger SELF_INSERTs, not being \n, for example, "World"

Surely we cannot wait for a period of unkown m to determine whether the \r at n+1 should have triggered ACCEPT_LINE.

[hvesalai]

Consider above with this user over a slow connection:

Time n: process is piped something that triggers SELF_INSERTs, such as "Hello"
Time n+1: process is piped \r
Time n+m: process is piped \n

Waiting for the \n here would be possible (using a short timeout > m), but that would mean that a dumb terminal would always wait after a \r to see if there is a \n following it. Sometimes it could be available, sometimes not. The bottom line is that it would in any case slow down the process.

[hvesalai]

@gnodet this PR is actually more important for me. I think it should be quite clean, so maybe you could merge this now and I can continue with the other PR as needed.

[gnodet]

Looks good

[gnodet]

@hvesalai Unit tests are failing...

[hvesalai]

@gnodet yes I know. See note here #874 (comment)

I wanted to get your feedback on how to fix this.

[gnodet]

@gnodet yes I know. See note here #874 (comment)

I wanted to get your feedback on how to fix this.

Something like the following should work:

diff --git a/reader/src/test/java/org/jline/terminal/impl/AbstractWindowsTerminalTest.java b/reader/src/test/java/org/jline/terminal/impl/AbstractWindowsTerminalTest.java
index f3dee5e6..85c9f30a 100644
--- a/reader/src/test/java/org/jline/terminal/impl/AbstractWindowsTerminalTest.java
+++ b/reader/src/test/java/org/jline/terminal/impl/AbstractWindowsTerminalTest.java
@@ -87,7 +87,7 @@ public class AbstractWindowsTerminalTest {
                     null,
                     new AnsiWriter(new BufferedWriter(sw)),
                     "name",
-                    AbstractWindowsTerminal.TYPE_DUMB,
+                    "windows",
                     Charset.defaultCharset(),
                     false,
                     SignalHandler.SIG_DFL,
(END)

[hvesalai]

@gnodet that fixes it, thank you! Pushed the fix to this PR.