On 04/23/2018 11:38 AM, Ian Jackson wrote:
> You are right. Perhaps my testing was inadequate. I wrote this a
> long while ago, and if there was a syntax along these lines that DTRT
> in both bash and dash in my tests it is long gone. Starting de novo,
> the following code works for me:
>
> (echo >>config.log "
> funcs: ${FUNCNAME[*]}
> lines: ${BASH_LINENO[*]}
> files: ${BASH_SOURCE[*]}") 2>/dev/null ||:
That's still fork-heavy. You could do:
test -n "$BASH_VERSION" && eval '
echo >>config.log "
funcs: ${FUNCNAME[*]}
lines: ${BASH_LINENO[*]}
files: ${BASH_SOURCE[*]}"'
which avoids the fork, but remains silent on dash.
>
> With bash I get the expected information in config.log, which looks
> like this:
>
> funcs: do_compiler do_cc compile_prog cc_has_warning_flag main
> lines: 91 124 1720 1724 0
> files: ./configure ./configure ./configure ./configure ./configure
Is files: really useful information? The other two are (as it gives a
full stack trace), but if we aren't actively sourcing lots of other
files, seeing a bunch of ./configure doesn't add much.
>
> With dash the script runs but there is nothing from this segment in
> the log. Without the 2>/dev/null, it prints
> ./configure: 63: ./configure: Bad substitution
> so the syntax error is indeed being suprresed and ignored.
>
> The ( ) is necessary because syntax errors are not like set -e errors:
> they cause the shell process to exit.
See above - a well-quoted eval is sufficient to avoid a subshell.
--
Eric Blake, Principal Software Engineer
Red Hat, Inc. +1-919-301-3266
Virtualization: qemu.org | libvirt.org
signature.asc
Description: OpenPGP digital signature
