[docs] LibraryEvolution: Feedback from DaveA and co.

[jrose-apple]

• Clear up allowed changes for associated types.
• Modifying a fixed_contents struct's properties' accessors isn't inherently ABI-breaking, just behavior-changing like any other inlineable thing.
• Clarify what an "empty extension" is.

cc @dabrahams

[jrose-apple]

@swift-ci Please smoke test

[slavapestov]

I don't see how removing a default can be binary compatible.

Suppose version 2 adds an associated type with a default, and version 3 removes the default. A client compiled against version 1 will not have the associated type in its witness table.

[jrose-apple]

Oops, good point. Is it worth allowing the removal if the associated type has been around as long as the protocol, or should we just leave it out?

[jrose-apple]

Removed.

[jrose-apple]

@swift-ci Please smoke test

[jrose-apple]

@shahmishal Here's one where it failed to detect the hash mismatch for some reason: https://ci.swift.org/view/Pull%20Request/job/swift-PR-Linux-smoke-test/1595/

[shahmishal]

I might have to fetch/pull before checking sha

if git -C ${WORKSPACE}/swift cat-file -e ${ghprbActualCommit} 2> /dev/null ; then
   echo "--- Correct sha ${ghprbActualCommit}  ---" 
else    
   echo "--- Wrong sha ${ghprbActualCommit}  ---" 
   export new_ghprbActualCommit=$(git  -C ${WORKSPACE}/swift show-ref --hash refs/remotes/origin/pr/${ghprbPullId}/head)
   echo "ghprbActualCommit=${new_ghprbActualCommit}" > ${JOB_WORKSPACE}/re_build.properties
   echo "--- EXIT WRONG SHA ---"
   exit 1
fi

[shahmishal]

@swift-ci Please smoke test Linux