GNAT SJLJ Build project

07/07/19, modified 20/07/19

Getting a SJLJ build working is not going well. Although success seemed just around the corner 2 weeks ago, the whole project is bogged down and seemingly going nowhere. To help myself, I'm making a list of the foreseeable steps to take next and an estimation on when I hope to have a build done.

First, the current status. The builds of the cross compiler for x86-64 and aarch64 seem to work. But this only seems so, as the test I am using to see if aborting a thread works, does not work. Also, the next step, building the native compilers crashes. The crash has to do with a jump instruction somewhere in a file called "haifa_sched.c". First inspection of the "haifa_sched.c" file was depressing and looks like sphagetti.

The plan (left is failure, right is success);

plandot

Currently I expect each task to tak

e bout a day, with the hope this will improve.

Estimate to have this project done: 15-september-2019

Estimate to have this project done: 15-September-2019

Update (20-July-2019):

And so it goes, time flies while you are looking for ways on not working on a task you do not want to do yet.

plandot1

I've updated the plan graph and added some more items, now I need to find a way to make clear that;

  1. I'm first going to re-try to get a working test-case with the AdaCore binaries. All the ingredients are already available and have been reported in the forum to work, so this should not be hard. The difference with the previous attempt is that I'll document it.
  2. Next, given the working test-case, I'm going to work on getting a working ave1 cross compiler for intel.
  3. Next, given the working test-case + cross-compiler, I'm going to work on getting a working native compiler for intel.
  4. Next (not in the graph yet), given the working test-case + cross-compiler + native compiler for inter, I'm going to do step 2 and 3 for ARM.

Of course, my previous attempt was to go for the last point in step 4. If that had worked then it would have provided me with the shortest path to the finish. But once on the path it is very hard to simply "core dump"1 the whole thing and start from scratch.

  1. Yes I know the term, what it means and that this is the wrong usage. I call this humor, my only excuse being that I'm boring. []

One Response to “GNAT SJLJ Build project”

  1. It seems to me sinking the week or so into checking and perfecting this graph is worthwhile ; by mid July we should be in a better position to see if it's worth sinking your time till Autumn in chasing this or maybe regroup and see something else.

Leave a Reply to Mircea Popescu