Infinite loop on syntax errors

[nahoj]

When readable-traceback is loaded from an rc file, any syntax error (e.g. typing { or echo {) causes an infinite loop that freezes the shell.

Cause

The __flush function in readable-traceback.xsh (line 40) contains:

rep = lambda s: '' if '__amalgam__.py' in s or 'readable-traceback.xsh' in s else s+'\n'

xonsh compiles '__amalgam__.py' in s or ... as a subprocess command rather than a Python expression. So every call to __flush tries to execute __amalgam__.py as a shell command.

When a syntax error occurs → _print_exception → backtrace.hook → __flush → subprocess __amalgam__.py → command not found → _print_exception → loop.

However, this doesn't happen if xontrib load readable-traceback is run in interactive mode, because then it doesn't replace ptk_shell.print_exception.

Reproduction

Without the xontrib:

@ {

xonsh: For full traceback set: $XONSH_SHOW_TRACEBACK = True
  File "<stdin>", line 1
    {
    ^
SyntaxError: EOF in multi-line statement

With it:

 @ {

[infinite loop]
Traceback (most recent call last):
[nothing]

Tested on xonsh 0.22.8.