Another update, this time with good news! I'm writing this listening to music with a demo plugin running simultaneously on a mini2g (PP502x) built using gcc 4.9.4 (see g#2305)
There turned out to be two major (and independent) problems:
* arm corelock asm code was relying on constructs that, according to
GCC docs, "cannot be depended upon to work reliably and are not
supported".
* An odd quirk in the linker script caused overlapping sections with
binutils 2.21 or newer. (This overlap included things like various
mutexes. Cue much badness..)
I've already committed fixes for these, and the pile of more minor
issues I've found along the way. No regressions noted using the old
tooling.
Now there are still many build warnings in code that wasn't previously
touched by the mips, native, and hosted 4.9.4-based builds. Many (if
not most) are triggered by using -Wextra and and a change of default
optimization from -O to -Os.
So. I'm not comfortable committing this toolchain bump without wider
testing of gcc494 builds on sh, m68k and various other native arm
targets (especially other portalplayer-based units!)
Any volunteers?
- Solomon
--
Solomon Peachy pizza at shaftnet dot org (email&xmpp)
@pizza:shaftnet dot org (matrix)
High Springs, FL speachy (freenode)
signature.asc
Description: PGP signature
