In a few words, this PR:
(of course, github tangled with the commits ordering…)
duration in milliseconds, roughly of ~evaluating two particular lines of code, after some warm-up
I couldn't run my benchmark as is with the master of Ammonite (seems the
Once the CI is green here, I can post the detailed methodogy if that's of interest to anyone…
I'll probably merge that in the coming days, unless anyone opposes.
About the commented out failing test case, it seems unrelated to the core of this PR… More likely an unearthed corner case in TPrint (it only happens in 2.12, when class wrapping is enabled, the latter not being enabled by default).
The only point I'm not too sure about is whether this can mess with the terminal when one runs commands with
Compiler was refreshed after each line of code, after interp.preConfigureCompiler was called. With this commit, it's only refreshed once (as was originally intended I guess).
Have Ctrl-C not send SIGINT to us. Ctrl-C are then handled by terminal filters, like before (current input line is ignored, new prompt is printed below, accepting new input). This runs a command right before and after accepting input, which has some performance implications… This reverts some gains of com-lihaoyi#853.