remove config step auto::jit

[coke]

Looks like it hardcodes some values in the config, but doesn't do
much. Since we don't currently have a JIT, this config step can
probably be removed without any harm.

-- 
Will "Coke" Coleda

[coke]

This message has 0 attachment(s)

[jkeenan]

IIRC, when we ripped out the old, dysfunctional JIT, we considered removing this config step (and its associated tests as well). But some people felt we'd be getting the new JIT before long, so there was no harm in leaving it in.

Well, "before long" hasn't arrived yet, so I have no objection to proceeding with removal.

But perhaps we could get a GCI task out of this:

Evaluate the impact of removing configuration step auto::jit.

Task would be to (a) identify the impact of removing any of the values currently set by this configuration step -- hopefully demonstrating that their removal does no harm; (b) actually remove the config step, its tests, etc.

How does that sound?

kid51

[jkeenan]

But perhaps we could get a GCI task out of this:

See  this published GCI task.

[jkeenan]

diff of remove_auto_jit branch against its branch point

[jkeenan]

Created the remove_auto_jit to work on this ticket. Removed the configuration step, its associated test, associated mentions in documentation, options lists, etc.

I had to touch more files than I would have expected, largely because there are instances of EXEC_CAPABLE and JIT_CAPABLE in quite a few places in the code.

Please review the attached patch, which passed make test and make fulltest.

Thank you very much.

kid51

[jkeenan]

Replying to jkeenan:

Please review the attached patch, which passed make test and make fulltest.

On review, it may not have passed make fulltest. It may have failed make benchmark_tests. I'll investigate, but this shouldn't stop others from reviewing the patch.

kid51

[jkeenan]

Replying to jkeenan:

On review, it may not have passed make fulltest. It may have failed make benchmark_tests. I'll investigate, but this shouldn't stop others from reviewing the patch.

Never mind; couldn't replicate the failure.

[jkeenan]

Per discussion with cotto, merged this branch into trunk. The merge was messy, but everything should be okay now.

We noted that we still have t/op/jit.t and t/op/jitn.t is the distribution. However plobsing points out that "they were conceived as basic execution tests for the JIT, but they are applicable to anything that executes ops". So we'll leave them be.

Will keep ticket open for 2-3 days for comments, complaints.

Thank you very much.

kid51

[jkeenan]

No complaints; closing ticket.