Bug in install applet

Rob Landley rob at landley.net
Wed Feb 24 21:09:50 UTC 2010


On Wednesday 24 February 2010 12:18:49 Mikhail Gusarov wrote:
> Twas brillig at 09:21:33 24.02.2010 UTC-06 when rob at landley.net did gyre
> and gimble:
>
>  RL> You can't depend on the host strip to support your target, _or_ to
> reliably RL> complain when it doesn't.
>
> strip --info | grep || echo "Strip is too stupid".

And the "install" command (which would presumably be modified to always and 
forever call strip twice on the off chance it was cross compiling) knows which 
platform it's attempting to work on how?  Since "install-" itself isn't 
prefixed, I'm guessing we bloat the busybox version even more to first call 
"file" on the binary to identify the type, parse the output of that, adn then 
call strip --info and parse the output of _that_ in order to work around an 
existing lack of proper error checking in the Free Software Foundation's code?

And this is any of busybox's business why?

You're way into over-engineered territory.  Your best-case scenario is you 
manage to reliably break the build when people use install -s in cross compile 
situations with host toolchains that work for everything _else_, and what 
you're doing is working around a bug in a presumably cross-compile-aware 
package that actually isn't.  Your panacea is interesting but not actually a 
fix to the issue I raised.

I know, I know, "tl;dr".  No point arguing with you, you won't read it...

Rob

P.S.  Don't get me started on binflt.
-- 
Latency is more important than throughput. It's that simple. - Linus Torvalds


More information about the busybox mailing list