#4888 spurios output in 'make test'

[miguel sofer]

Running 'make test' on today's trunk tip shows (look right after chan.test)

Tests began at Tue Aug 09 18:35:22 ART 2011
append.test
appendComp.test
apply.test
assemble.test
assocd.test
async.test
autoMkindex.test
basic.test
binary.test
case.test
chan.test
'
chanio.test
clock.test

[Alexandre Ferrieux]

See comment about this in 2919042: it is test chan-4.6 which sets an output oefchar on stdout (the single quote).
It turns out that full finalization (you used -DPURIFY, right ?) forces the output eofchar to be written.

[Alexandre Ferrieux]

In 8.5, though Finalize was called, the closing of (std?) channels was not complete.
You can observe the same behavior in 8.5 by adding [close stdout] at the end.
So I'd say this is the correct behavior, bugs are behind us...

[miguel sofer]

So is it a bug in the test itself? Is it reported, or is anybody fixing it already?

[Alexandre Ferrieux]

Since it occurs only as a result of today's commit of full-finalize-on -DPURIFY, nobody has had yet a chance to fix it I guess...

The problem is that the semantics of full finalization are in a grey area -- no specs :P
Maybe we should work to bring back symmetry, so that even in a non-finalized case we get the same output; then we can fix the test.

[Alexandre Ferrieux]

Ah, I've found the (old) bug that has been snuffing the eofchar for ages: 653218 "Channels CloseProc() not called on close"

[Alexandre Ferrieux]

Hm, strike that one, the problem is different. Opening new bug 3389289 to keep track of it. Let's come back here when that knot is untangled.

[Alexandre Ferrieux]

Okay, just committed to 8.6 trunk (c6db0a87d1) what was really an unwanted side-effect from chan-4.6.
Continuiing the discussion about close/exit/eofchar in 3389289.