Would anyone be opposed to a new option to redirect hs-err files to stderr/out?

[David Holmes]

Hi Thomas,

I commented on the CSR before I saw this email thread. I really hoped we 
could enhance ErrorFile rather than adding yet-another-flag.

Do we seriously think anyone would realistically use stderr and stdout 
as the names of custom error files? I'd be okay addressing that with a 
release note (though if you wanted to backport this it is more of an issue).

That said, given we already have DisplayVMOutputToStderr/Stdout then 
something like -XX:+ErrorFileToStderr/Stdout may be the better choice.

That aside I see this as a flaw with the environments you're running in. 
If an application can produce files then the container environment 
should provide a way to handle that properly. Writing to stdout/stderr 
is really just a hack - and you may get a real mess of interleaved 
output. :(

Cheers,
David

On 19/03/2019 2:16 am, Thomas Stüfe wrote:
> On Mon, Mar 18, 2019 at 4:38 PM Remi Forax <forax at univ-mlv.fr> wrote:
> 
>> Hi Thomas,
>>
>> ----- Mail original -----
>>> De: "Thomas Stüfe" <thomas.stuefe at gmail.com>
>>> À: "Aleksey Shipilev" <shade at redhat.com>
>>> Cc: "hotspot-runtime-dev" <hotspot-runtime-dev at openjdk.java.net>
>>> Envoyé: Lundi 18 Mars 2019 15:39:00
>>> Objet: Re: Would anyone be opposed to a new option to redirect hs-err
>> files to stderr/out?
>>
>>> On Mon, Mar 18, 2019 at 2:50 PM Aleksey Shipilev <shade at redhat.com>
>> wrote:
>>>
>>>> On 3/18/19 2:43 PM, Thomas Stüfe wrote:
>>>>> We have the -XX:ErrorFile option to overwrite the default path and
>> file
>>>>> name for the hs-err file. I would like to add the option of
>> specifiying a
>>>>> special value which would mean "stdout" or "stderr", respectivly.
>>>>>
>>>>> E.g. -XX:ErrorFile=STDERR or -XX:ErrorFile=STDOUT .
>>>>>
>>>>> Any objections? If not, I would draft a CSR and implement this.
>>>>
>>>> That would be very welcome.
>>
>> yes, i've the same issue when using TravisCI
>>
>>>>
>>>> It would be the bikeshedding exercise to come up with the special name
>>>> that would not clash with
>>>> potential real user-specified path.
>>>
>>>
>>> :( I was hoping... no, you are right.
>>>
>>>
>>>> On Linux, I'd expect -XX:ErrorFile=/dev/stdout to work -- maybe
>>>> start from making sure that works reliably?
>>
>> -XX:ErrorFileRedirect=stderr is better IMO
>>
>>
> Thanks. I like this better.
> 
> I opened a CSR for this. https://bugs.openjdk.java.net/browse/JDK-8220787
> 
> Lets see how this is received.
> 
> ..Thomas
> 
> 
>>>>
>>>>
>>> It fails now with EEXISTS, since we open the file as new. Changing that
>> to
>>> accept O_APPEND opens a different can of worms: do you want people to
>>> append post-mortems to existing files?
>>>
>>> Also, /dev/stdout is not supported on all platforms, eg not on Windows
>> and
>>> our beloved AIX.
>>>
>>> A pragmatic approach but not as elegant as reusing -XX:ErrorFile would
>>> adding a new option, or options, -XX:RedirectHsErrFileTo(Stdout|Stderr).
>>>
>>> In that case we would have to think about what to do if both are
>> specified.
>>>
>>> ..Thomas
>>>
>>>
>>>
>>>> -Aleksey
>>>>
>>