I don't understand what I'm missing here:
$ echo 'testing' | sed -E 's/([a-z]*)ing/\1ing/g'
ing
I would expect the output to be testing again, since \1 should be test? That input seems to have been swallowed - i.e. the group matched it - but why is \1 not spitting it back out?
I am on macOS 10.12.2; using xonsh shell and GNU sed v4.3.
It turns out xonsh (or Python) is swallowing the \1, so sed actually sees 's/([a-z]*)ing/ing/g and its output is of course correct for that input.
I've opened an issue here about it, but the workaround is to use a Python raw string:
$ echo 'testing' | sed -E r's/([a-z]*)ing/\1ing/g'
testing
This is preferable to escaping (\\1) since it would error in a POSIX shell rather than continue with the undesired \1.
However, with thanks to @adqm, double-quoting and also escaping the backslash is portable between xonsh and bash:
$ echo 'testing' | sed -E "s/([a-z]*)ing/\1ing/g"
echo '\1' | hexdump: 0000000 01 0a [line end] 0000002 (under xonsh)
\1 is being interpreted as an octal escape for a byte 01, which sed will happily accept as part of the replacement. \\1 should work too then.
Commented
Jan 15, 2017 at 5:52
\\1 is fine, less desirable because it doesn't error in POSIX shells (and continues with a literal \1). Maintainer pointed out in Github issue though that double-quoting and escaping ('s//\\1/') will work in both xonsh and bash.