The release notes for the latest z/OS C/C++ compiler are interesting. When I was at IBM they were working on “clangtana”, a clang frontend melded with the legacy TOBY backend. This really surprised me, but was consistent with the fact that the IBM compiler guys kept saying that they were continually losing their internal funding — that project was a clever way to do more with less resources. I think they’d made the clangtana switch for zLinux by the time I left, with AIX to follow once they had resolved some ABI incompatibility issues. At the time, I didn’t know (nor care) about the status of that project on z/OS.
Well, years later, it looks like they’ve now switched to a clang based compiler frontend on z/OS too. This major change appears to have a number of side effects that I can imagine will be undesirable to existing mainframe customers:
- Compiler now requires POSIX(ON) and Unix System Services. No more compilation using JCL.
- Compiler support for 31-bit applications appears to be dropped (64-bit only!)
- Support for C, FASTLINK, and OS linkage conventions has been dropped (XPLINK only.)
- Only ibm-1047 is supported for both source and runtime character set encoding.
- C89 support appears to have been dropped.
- Hex floating support has been dropped.
- No decimal floating point support.
- SIMD support isn’t implemented.
- Metal C support has been dropped.
i.e. if you want C++14, you have to be willing to give up a lot to get it. They must be using an older clang, because this “new” compiler doesn’t include C++17 support. I’m surprised that they didn’t even manage multiple character set support for this first compiler release.
It is interesting that they’ve also dropped IPA and PDF support, and that the optimization options have changed. Does that mean that they’ve actually not only dropped the old Montana frontend, but also gutted the whole backend, switching to clang exclusively?