[blfs-support] OpenJDK

[Paul Rogers]

This comes from a PM with Pierre.

I'm not entirely comfortable to present running the test suite as
"optional". I'd be happier if it was presented as one of the
"strongly recommended", such as running the tests on glibc. The
potential for a catastrophic flaw is just a little too great.

Well maybe. I suggest you discuss that on the list. I can't decide
alone anyway.

OK, I will.

So, can we have some discussion about this and come to a consensus? I
wouldn't even consider NOT running the test quite on OpenJDK to validate
my build. It's not "Optional" in my view, and I'd never suggest it is.
--
Paul Rogers
***@fastmail.fm
Rogers' Second Law: "Everything you do communicates."
(I do not personally endorse any additions after this line. TANSTAAFL
:-)

[Fernando de Oliveira]

Post by Paul Rogers
This comes from a PM with Pierre.

I'm not entirely comfortable to present running the test suite as
"optional". I'd be happier if it was presented as one of the
"strongly recommended", such as running the tests on glibc. The
potential for a catastrophic flaw is just a little too great.

What would be a catastrophic flaw? How the tests could prevent it?

I see in the site:

jdk8u60-b12 Passed Failed* Not Run** Error Summary Diff
jdk 4787 18 822 1 5628 7
hotspot 627 34 13 4 678 2
langtools 3078 0 0 0 3078 0

jdk8u60-b11 Passed Failed* Not Run** Error Summary Diff
jdk 4787 17 819 0 5623 1
hotspot 625 35 13 5 678 2
langtools 3078 0 0 0 3078 0

There is a relatively small number of failures, but the numbers are not
so small.

Post by Paul Rogers

Well maybe. I suggest you discuss that on the list. I can't decide
alone anyway.

OK, I will.

So, can we have some discussion about this and come to a consensus? I
wouldn't even consider NOT running the test quite on OpenJDK to validate
my build. It's not "Optional" in my view, and I'd never suggest it is.

I agree with Pierre, the instructions in the book have been much improved.

The comments "the test infrastructure of OpenJDK is a work in progress"
and "testing ... involves several steps" seem appropriate for an
optional part of the instructions.





To Pierre:

Following remarks are consequences of reading to reply to this post. The
page, as I said, has been much improved, since you started taking care
of it. Congratulations.

Following sentences have typo(s?) and perhaps could also be improved (it
may have been partially or totally written by me, so you could blame me):

{{{
There are a few failures, the number of which depends on various
conditions, like whether the computer is connected to network. Also,
some test may timeout if the machine is under load.
}}}

I don't know if "There are a few failures" is correct, "are" plural, "a"
singular.

s/some test/&s/ (learned with you)

Spell check does not recognize "timeout", but "time-out" or "time out".
Sometimes, I don't agree with it, but you decide.

[Paul Rogers]

Post by Fernando de Oliveira
What would be a catastrophic flaw? How the tests could prevent it?

I experienced a build/test failure making .31, and it showed up in the
test results.

The point is rather whether not testing is viable. If one doesn't test
one has no confidence at all that it has been made correctly. (I grant
that just looking at the tables and comparing gross numbers does little
to create much confidence. But not doing so cannot be better!) This
isn't ed we're making here. It's a HUGE package with a known history of
security vulnerabilities. I imagine there are any number of possibile
ways for an incorrect compilation to compound those vulnerabilities.

Post by Fernando de Oliveira
I agree with Pierre, the instructions in the book have been much
improved.

No argument there, and I think it will continue to improve.

Post by Fernando de Oliveira
The comments "the test infrastructure of OpenJDK is a work in
progress" and "testing ... involves several steps" seem appropriate
for an optional part of the instructions.

Of course it is, and it shouldn't be removed. The question remains:
with such a large project, is it right suggest avoiding running the
tests for 110 SBU's can be avoided without consequences.

Post by Fernando de Oliveira
{{{ There are a few failures, the number of which depends on various
conditions, like whether the computer is connected to network. Also,
some test may timeout if the machine is under load. }}}
I don't know if "There are a few failures" is correct, "are" plural,
"a" singular.
s/some test/&s/ (learned with you)

Some tests, plural, DO timeout, even on an i7-940, and the consequences
of not having networking running are terrible--like 57 orphaned JVM's
never cleaned-up and left running! That statement needs strengthening.
"Networking must be running."
--
Paul Rogers
***@fastmail.fm
Rogers' Second Law: "Everything you do communicates."
(I do not personally endorse any additions after this line. TANSTAAFL :-)