msp430-elf toolchain

[JeremyGrosser]

Before this toolchain is useful to anyone, the msp430-elf target needs to be added to gprconfig_kb in order for gprbuild to find it. I did that in this commit, based on gprconfig_kb's master branch. GNAT-FSF-builds uses the 22.0.0 release of gprconfig_kb, so we'll either need to backport this patch or switch to a newer version.

I ported TI's blink demo to SPARK/Ada, using bare_runtime: https://github.com/JeremyGrosser/msp430test/blob/master/src/msp430test.adb

The main procedure passes gnatprove verification, but bare_runtime does not, so analysis of those units is skipped.

[JeremyGrosser]

I've generated linux64 binaries for this toolchain and setup a separate Alire index repo to point to them. Maybe it makes sense to just maintain these things separately to reduce @mosteo's workload. https://github.com/JeremyGrosser/alire-index-msp430

[Fabien-Chouteau]

Hi Jeremy thanks for the contribution. To be honest I won't have much time to look at this in the short term.

About the separate index, that's a good short term solution. But for the mid to long term it's better to have all the toolchain stuff built from the same place and available in the same index.

[JeremyGrosser]

Makes sense, thanks for taking a look!

[JeremyGrosser]

I tried porting bb-runtimes to this architecture, but I ran into some trouble with libgnat. It uses Unsigned to store the mantissa of IEEE Float. There's an assertion that checks that 'Machine_Mantissa <= Uns'Size, which fails. The implementation of Machine_Mantissa has a minimum of 24 bits, so this obviously won't work on a 16-bit machine.

Patching libgnat or removing Float support is a bit further than I want to go with this project, so I'm abandoning this PR. TI themselves have announced some upcoming chips with the MSP430 peripherals attached to a Cortex-M0+, so I think the days are numbered for this architecture anyway.

[Fabien-Chouteau]

If you can find a solutionbare_runtime for this problem in I am interested.