RFR: JDK-8248158 Configure fails with autoconf not found even though it's installed

[Simon Tooke]

(Disclaimer: I am not a reviewer, so this is an opinion, not a review)

I have tested this on Windows and it built without issue, although the 
submit repo should be the final gate.  I'd also like to add my void to 
simply redefining 'WHICH' as it leads to less changes in the source 
code, which would make life easier should this be backported to 11u 
and/or 8u.

-Simon

On 2020-07-02 4:22 a.m., Galder Zamarreno wrote:
> On Thu, Jul 2, 2020 at 12:37 AM Magnus Ihse Bursie <
> magnus.ihse.bursie at oracle.com> wrote:
>
>>
>> On 2020-07-01 12:05, Galder Zamarreno wrote:
>>> Using `which` to check whether commands exist can result in confusing
>>> errors when `which` itself is not installed in the system. This is the
>> case
>>> with `autoconf`, where if `autoconf` is present but `which` isn't, the
>>> build system says that `autoconf` is missing, when in reality it is
>> `which`
>>> which is missing. The fix switches autoconf uses of `which` for `type -p`
>>> instead, which is a Bash built-in command.
>>>
>>> I've tested the fix with a fedora docker container that had `autoconf`
>>> installed but `which`. When using `type -p` it correctly detects
>> `autoconf`
>>> installed and eventually fails saying that `which` is not installed,
>> which
>>> is the expected behaviour.
>>>
>>> `which` is still in use in make/autoconf/util_windows.m4. A possible
>> future
>>> improvement would be to see if `which` use there could be replaced as
>> well.
>>> Eventually, when no `which` uses remain, the presence check for `which`
>>> could be removed.
>> I agree that we should replace "which" with "type -p" everywhere. The
>> best way to do this is probably to replace the value of $WHICH with
>> "type -p". It's a bash built-in, so we can count on its presence. If you
>> want to fix that as part of this bug, I'm ok with it, otherwise we
>> should open a new bug to track this. I think there is also one or two
>> instances of "command" recently added as (better, but not as good as
>> "type -p") replacements for which.
>>
> I discovered one place in util.m4 where command was being used.
>
> There are other places outside of make/ where command is used but I feel
> those are a bit out of scope here.
>
> My main objective is that from a configure perspective, we'd try to reduce
> the number of dependencies needed to build things.
>
> I'll send an updated webrev shortly.
>
>
>> /Magnus
>>> Bug: https://bugs.openjdk.java.net/browse/JDK-8248158
>>> WebRev:
>> http://cr.openjdk.java.net/~sgehwolf/webrevs/JDK-8248158/01/webrev/
>>> Galder
>>