MCCAS fixes on next.

[rastogishubham]

This commit has 3 minor, but important fixes

1. Ensure that MCLEBFragmentRef::create uses the passed-in FragmentsContents to create the cas object.

2. Ensure that MCCVInlineLineTableFragmentRef::create uses the passed-in FragmentsContents to create the cas object.

3. Make FT_CVDefRange a mergeable fragment, as all other non-encoded fragments are mergeable.

rdar://159458401

rdar://159216364

[rastogishubham]

@swift-ci please test llvm

[cachemeifyoucan]

I can see LEB is covered by existing test case. Can you add test for other changes?

[rastogishubham]

I can see LEB is covered by existing test case. Can you add test for other changes?

I don't know how to create a testcase for MCCVInlineLineTableFragmentRef, how do I write an .ll file that will guarantee that the MCAssembler will produce a MCCVInlineLineTableFragment with both contents and variable contents?

As for making FT_CVDefRange a mergeable fragment, it is not a bug fix, it is just an NFC change for consistency

[cachemeifyoucan]

I don't know how to create a testcase for MCCVInlineLineTableFragmentRef, how do I write an .ll file that will guarantee that the MCAssembler will produce a MCCVInlineLineTableFragment with both contents and variable contents?

You want to find that out. We need to gradually increase the coverage of MCCAS in lit testing.

[ojhunt]

I don't know how to create a testcase for MCCVInlineLineTableFragmentRef, how do I write an .ll file that will guarantee that the MCAssembler will produce a MCCVInlineLineTableFragment with both contents and variable contents?

You want to find that out. We need to gradually increase the coverage of MCCAS in lit testing.

This is breaking existing tests for me, so it would seem reasonable to land the fix as it stands while working out how to add additional tests, or I guess to revert the original PR :-/

[rastogishubham]

@cachemeifyoucan the only way to figure it out would be to get an MCCAS round-trip verification error where a CVInlineLineTableFragment contains both, variable content, and fixed content, and ideally a relocation addend too. I don't think anyone can write an ll file by hand to make that happen