> because you want avoid 10 times bigger binary? (alpine linux's
> apk-tools is 90k dynamic and 900k static due to libcrypto for
> hashing/signing)

 So what? libcrypto obviously wasn't designed for embedded
environments. In fact, it's probably the most bloated free crypto
library around. If the Alpine Linux guys need a smaller static
executable, they can use a smaller crypto library, that does not
pull the whole plate of spaghetti plate, glibc-style, when you
yank at a single noodle.

 Do not blame extra bloat on static linking. Blame it on people
who can't design lean software.


> Still i think beeing able to set ABI_VERSION is a good idea.
> (...)
> But the price is that extra bloat for libcrypto, libz and uclibc dupe
> code in there. I do understand that some people are willing to pay
> that price. No problem. Current git is good. Everyone is happy. Lets
> move on.

 That's fine, as long as the support for ABI_VERSION does not impact
code that does not use it. If statically linked code is somehow made less
efficient by adding dynamic ABI_VERSION support, not everyone's happy.
If the libc.so dynamic code is somehow made bigger by adding dynamic
ABI_VERSION support and users don't need it, not everyone's happy.

 In embedded environments, RAM is expensive, but storage memory is
getting cheaper by the second. Cellphones embed full web browsers with
Java and Macromedia Flash support.
 In most cases, bigger binaries for a package management suite will
be *a lot* less of a problem than dynamic library upgrading nightmares.
Vendors of embedded products want the upgrade system to just work. If
it means adding 32 MB of storage memory and eliminating all risk of
failure, they will do it.

 Support for upgrading the library does not belong to the libc. It
belongs to the upgrade system. The libc should just do its work;
thinking outside of itself, going meta, is going the way of the glibc.
Let package manager designers handle the upgrading of the libc. Package
managers are where it's at.

-- 
 Laurent
_______________________________________________
uClibc mailing list
[email protected]
http://lists.busybox.net/mailman/listinfo/uclibc

Reply via email to