You are viewing a plain text version of this content. The canonical link for it is here.
Posted to dev@lucene.apache.org by Erick Erickson <er...@gmail.com> on 2015/08/25 18:23:04 UTC
Should we EOL support for 32 bit systems?
I have no real skin in this game, but I thought it worth asking after
Uwe's recent e-mail about disabling 32bits with -server tests.
I guess it boils down to "who is using 32-bit versions?". Are we
spending time/energy supporting a configuration that is not useful to
enough people to merit the effort?
I'm perfectly content if the response is "That's a really stupid
question to ask, of course we must continue to support 32-bit OSs".
Although some evidence would be nice ;)
It's just that nobody I work with is running 32 bit OS's. Whether
that's just my limited exposure to people running small systems is
certainly a valid question.
If we _do_ decide to drop support for 32 bit systems, what's the right
version? 6.0?
Random thoughts on a slow Tuesday morning....
Erick
---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscribe@lucene.apache.org
For additional commands, e-mail: dev-help@lucene.apache.org
RE: Should we EOL support for 32 bit systems?
Posted by Uwe Schindler <uw...@thetaphi.de>.
Hi,
the time needed to keep Jenkins running in 32 bits is not so time consuming, it is just downloading and unzipping two instead of one tarball. Config changes are search/replace on one file using Notepad++. The time consuming part is keeping track of the bugs and communicate them. But as said before, the bug may only happen on 32 bit, but of course this may not be the real reason for the bug. There must be something wrong in Hotspot Compiler that just triggers when you use 32 bits and enable server + tiered compilation. The issue @ OpenJDK was already opened; will be visible soon.
Dawid and I had done some conversation this morning and I stopped Policeman Jenkins to have full testing capacity to run "ant beast" on full Lucene Core (ant beast -Dbeast.iters=100 -Dargs="..."). And the bug reproduced *always* on 32 bits + server VM + tiered compiler, while it never reproduced in client VM or any 64 bits VM. It was harder to reproduce without tiered compilation. <-- that was the outcome of my morning exercise with Hotspot :-) But with that information it should be easier to maybe find the bug (for somebody working on Hotspot). But in any case, the same could have hapenend with 64 bits, too (and I am glad that we tested 32 bits, otherwise we could have a similar problem like with 7u40 after release of Java 9)!
FYI, I am planning to add another VirtualBOX VM to Policeman using OpenSolaris. Unfortunately, this is not SPARC (I only have a x64 CPU), but a still often used platform. I just have to assign an IPv4 address to the VM, IPv6 is already available.
We should maybe also try to get one of those ASF Jenkins VMs with PowerPC CPU :-) I know they got some from Yahoo (I think).
Uwe
-----
Uwe Schindler
H.-H.-Meier-Allee 63, D-28213 Bremen
http://www.thetaphi.de
eMail: uwe@thetaphi.de
> -----Original Message-----
> From: Erick Erickson [mailto:erickerickson@gmail.com]
> Sent: Tuesday, August 25, 2015 6:47 PM
> To: dev@lucene.apache.org
> Subject: Re: Should we EOL support for 32 bit systems?
>
> Fair enough, just wanted to be sure we weren't making extra work for
> ourselves.
> Well, actually extra work for you since you seem to be the one who interacts
> with the compiler folks the most ;).
>
> On Tue, Aug 25, 2015 at 9:43 AM, Uwe Schindler <uw...@thetaphi.de> wrote:
> > Hi,
> >
> > From a Java point of view, there is no real difference to not support 32 bit
> versions. The bug with JDK 9 are just bugs that could also have happened
> with 64 bit versions. It is just easier to trigger this bug with 32 bits, but I am
> almost sure the underlying bug also affect 64 bits.
> >
> > So why should we no longer support all platforms Java supports? Bitness
> does not matter for our code? Should we then also no longer support Sparc,
> PowerPC, or ARM platform?
> > -1 to add arbitrary restrictions on our runtime environment. If we
> > want this, we should disallow all platforms we don't test on and of
> > course also al processor variants we don't test on! :-)
> >
> > Uwe
> >
> > -----
> > Uwe Schindler
> > H.-H.-Meier-Allee 63, D-28213 Bremen
> > http://www.thetaphi.de
> > eMail: uwe@thetaphi.de
> >
> >
> >> -----Original Message-----
> >> From: Erick Erickson [mailto:erickerickson@gmail.com]
> >> Sent: Tuesday, August 25, 2015 6:23 PM
> >> To: dev@lucene.apache.org
> >> Subject: Should we EOL support for 32 bit systems?
> >>
> >> I have no real skin in this game, but I thought it worth asking after
> >> Uwe's recent e-mail about disabling 32bits with -server tests.
> >>
> >> I guess it boils down to "who is using 32-bit versions?". Are we
> >> spending time/energy supporting a configuration that is not useful to
> >> enough people to merit the effort?
> >>
> >> I'm perfectly content if the response is "That's a really stupid
> >> question to ask, of course we must continue to support 32-bit OSs".
> >> Although some evidence would be nice ;)
> >>
> >> It's just that nobody I work with is running 32 bit OS's. Whether
> >> that's just my limited exposure to people running small systems is
> >> certainly a valid question.
> >>
> >> If we _do_ decide to drop support for 32 bit systems, what's the
> >> right version? 6.0?
> >>
> >> Random thoughts on a slow Tuesday morning....
> >>
> >> Erick
> >>
> >> ---------------------------------------------------------------------
> >> To unsubscribe, e-mail: dev-unsubscribe@lucene.apache.org For
> >> additional commands, e-mail: dev-help@lucene.apache.org
> >
> >
> > ---------------------------------------------------------------------
> > To unsubscribe, e-mail: dev-unsubscribe@lucene.apache.org For
> > additional commands, e-mail: dev-help@lucene.apache.org
> >
>
> ---------------------------------------------------------------------
> To unsubscribe, e-mail: dev-unsubscribe@lucene.apache.org For additional
> commands, e-mail: dev-help@lucene.apache.org
---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscribe@lucene.apache.org
For additional commands, e-mail: dev-help@lucene.apache.org
Re: Should we EOL support for 32 bit systems?
Posted by Dawid Weiss <da...@gmail.com>.
Plus, it is fun as well. A bit like finding these:
https://goo.gl/CQZPYh
Dawid
On Tue, Aug 25, 2015 at 11:37 PM, Jan Høydahl <ja...@cominvent.com> wrote:
> Another reason to keep 32bit support is that people who already have installed Java JRE from their browser tend to have the 32bit version (even if on 64bit OS). So they will be able to test-run Solr without re-installing Java.
>
> --
> Jan Høydahl, search solution architect
> Cominvent AS - www.cominvent.com
>
>> 25. aug. 2015 kl. 18.47 skrev Erick Erickson <er...@gmail.com>:
>>
>> Fair enough, just wanted to be sure we weren't making extra work for ourselves.
>> Well, actually extra work for you since you seem to be the one who interacts
>> with the compiler folks the most ;).
>>
>> On Tue, Aug 25, 2015 at 9:43 AM, Uwe Schindler <uw...@thetaphi.de> wrote:
>>> Hi,
>>>
>>> From a Java point of view, there is no real difference to not support 32 bit versions. The bug with JDK 9 are just bugs that could also have happened with 64 bit versions. It is just easier to trigger this bug with 32 bits, but I am almost sure the underlying bug also affect 64 bits.
>>>
>>> So why should we no longer support all platforms Java supports? Bitness does not matter for our code? Should we then also no longer support Sparc, PowerPC, or ARM platform?
>>> -1 to add arbitrary restrictions on our runtime environment. If we want this, we should disallow all platforms we don't test on and of course also al processor variants we don't test on! :-)
>>>
>>> Uwe
>>>
>>> -----
>>> Uwe Schindler
>>> H.-H.-Meier-Allee 63, D-28213 Bremen
>>> http://www.thetaphi.de
>>> eMail: uwe@thetaphi.de
>>>
>>>
>>>> -----Original Message-----
>>>> From: Erick Erickson [mailto:erickerickson@gmail.com]
>>>> Sent: Tuesday, August 25, 2015 6:23 PM
>>>> To: dev@lucene.apache.org
>>>> Subject: Should we EOL support for 32 bit systems?
>>>>
>>>> I have no real skin in this game, but I thought it worth asking after Uwe's
>>>> recent e-mail about disabling 32bits with -server tests.
>>>>
>>>> I guess it boils down to "who is using 32-bit versions?". Are we spending
>>>> time/energy supporting a configuration that is not useful to enough people
>>>> to merit the effort?
>>>>
>>>> I'm perfectly content if the response is "That's a really stupid question to ask,
>>>> of course we must continue to support 32-bit OSs".
>>>> Although some evidence would be nice ;)
>>>>
>>>> It's just that nobody I work with is running 32 bit OS's. Whether that's just my
>>>> limited exposure to people running small systems is certainly a valid
>>>> question.
>>>>
>>>> If we _do_ decide to drop support for 32 bit systems, what's the right
>>>> version? 6.0?
>>>>
>>>> Random thoughts on a slow Tuesday morning....
>>>>
>>>> Erick
>>>>
>>>> ---------------------------------------------------------------------
>>>> To unsubscribe, e-mail: dev-unsubscribe@lucene.apache.org For additional
>>>> commands, e-mail: dev-help@lucene.apache.org
>>>
>>>
>>> ---------------------------------------------------------------------
>>> To unsubscribe, e-mail: dev-unsubscribe@lucene.apache.org
>>> For additional commands, e-mail: dev-help@lucene.apache.org
>>>
>>
>> ---------------------------------------------------------------------
>> To unsubscribe, e-mail: dev-unsubscribe@lucene.apache.org
>> For additional commands, e-mail: dev-help@lucene.apache.org
>>
>
>
> ---------------------------------------------------------------------
> To unsubscribe, e-mail: dev-unsubscribe@lucene.apache.org
> For additional commands, e-mail: dev-help@lucene.apache.org
>
---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscribe@lucene.apache.org
For additional commands, e-mail: dev-help@lucene.apache.org
Re: Should we EOL support for 32 bit systems?
Posted by Jan Høydahl <ja...@cominvent.com>.
Another reason to keep 32bit support is that people who already have installed Java JRE from their browser tend to have the 32bit version (even if on 64bit OS). So they will be able to test-run Solr without re-installing Java.
--
Jan Høydahl, search solution architect
Cominvent AS - www.cominvent.com
> 25. aug. 2015 kl. 18.47 skrev Erick Erickson <er...@gmail.com>:
>
> Fair enough, just wanted to be sure we weren't making extra work for ourselves.
> Well, actually extra work for you since you seem to be the one who interacts
> with the compiler folks the most ;).
>
> On Tue, Aug 25, 2015 at 9:43 AM, Uwe Schindler <uw...@thetaphi.de> wrote:
>> Hi,
>>
>> From a Java point of view, there is no real difference to not support 32 bit versions. The bug with JDK 9 are just bugs that could also have happened with 64 bit versions. It is just easier to trigger this bug with 32 bits, but I am almost sure the underlying bug also affect 64 bits.
>>
>> So why should we no longer support all platforms Java supports? Bitness does not matter for our code? Should we then also no longer support Sparc, PowerPC, or ARM platform?
>> -1 to add arbitrary restrictions on our runtime environment. If we want this, we should disallow all platforms we don't test on and of course also al processor variants we don't test on! :-)
>>
>> Uwe
>>
>> -----
>> Uwe Schindler
>> H.-H.-Meier-Allee 63, D-28213 Bremen
>> http://www.thetaphi.de
>> eMail: uwe@thetaphi.de
>>
>>
>>> -----Original Message-----
>>> From: Erick Erickson [mailto:erickerickson@gmail.com]
>>> Sent: Tuesday, August 25, 2015 6:23 PM
>>> To: dev@lucene.apache.org
>>> Subject: Should we EOL support for 32 bit systems?
>>>
>>> I have no real skin in this game, but I thought it worth asking after Uwe's
>>> recent e-mail about disabling 32bits with -server tests.
>>>
>>> I guess it boils down to "who is using 32-bit versions?". Are we spending
>>> time/energy supporting a configuration that is not useful to enough people
>>> to merit the effort?
>>>
>>> I'm perfectly content if the response is "That's a really stupid question to ask,
>>> of course we must continue to support 32-bit OSs".
>>> Although some evidence would be nice ;)
>>>
>>> It's just that nobody I work with is running 32 bit OS's. Whether that's just my
>>> limited exposure to people running small systems is certainly a valid
>>> question.
>>>
>>> If we _do_ decide to drop support for 32 bit systems, what's the right
>>> version? 6.0?
>>>
>>> Random thoughts on a slow Tuesday morning....
>>>
>>> Erick
>>>
>>> ---------------------------------------------------------------------
>>> To unsubscribe, e-mail: dev-unsubscribe@lucene.apache.org For additional
>>> commands, e-mail: dev-help@lucene.apache.org
>>
>>
>> ---------------------------------------------------------------------
>> To unsubscribe, e-mail: dev-unsubscribe@lucene.apache.org
>> For additional commands, e-mail: dev-help@lucene.apache.org
>>
>
> ---------------------------------------------------------------------
> To unsubscribe, e-mail: dev-unsubscribe@lucene.apache.org
> For additional commands, e-mail: dev-help@lucene.apache.org
>
---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscribe@lucene.apache.org
For additional commands, e-mail: dev-help@lucene.apache.org
Re: Should we EOL support for 32 bit systems?
Posted by Erick Erickson <er...@gmail.com>.
Fair enough, just wanted to be sure we weren't making extra work for ourselves.
Well, actually extra work for you since you seem to be the one who interacts
with the compiler folks the most ;).
On Tue, Aug 25, 2015 at 9:43 AM, Uwe Schindler <uw...@thetaphi.de> wrote:
> Hi,
>
> From a Java point of view, there is no real difference to not support 32 bit versions. The bug with JDK 9 are just bugs that could also have happened with 64 bit versions. It is just easier to trigger this bug with 32 bits, but I am almost sure the underlying bug also affect 64 bits.
>
> So why should we no longer support all platforms Java supports? Bitness does not matter for our code? Should we then also no longer support Sparc, PowerPC, or ARM platform?
> -1 to add arbitrary restrictions on our runtime environment. If we want this, we should disallow all platforms we don't test on and of course also al processor variants we don't test on! :-)
>
> Uwe
>
> -----
> Uwe Schindler
> H.-H.-Meier-Allee 63, D-28213 Bremen
> http://www.thetaphi.de
> eMail: uwe@thetaphi.de
>
>
>> -----Original Message-----
>> From: Erick Erickson [mailto:erickerickson@gmail.com]
>> Sent: Tuesday, August 25, 2015 6:23 PM
>> To: dev@lucene.apache.org
>> Subject: Should we EOL support for 32 bit systems?
>>
>> I have no real skin in this game, but I thought it worth asking after Uwe's
>> recent e-mail about disabling 32bits with -server tests.
>>
>> I guess it boils down to "who is using 32-bit versions?". Are we spending
>> time/energy supporting a configuration that is not useful to enough people
>> to merit the effort?
>>
>> I'm perfectly content if the response is "That's a really stupid question to ask,
>> of course we must continue to support 32-bit OSs".
>> Although some evidence would be nice ;)
>>
>> It's just that nobody I work with is running 32 bit OS's. Whether that's just my
>> limited exposure to people running small systems is certainly a valid
>> question.
>>
>> If we _do_ decide to drop support for 32 bit systems, what's the right
>> version? 6.0?
>>
>> Random thoughts on a slow Tuesday morning....
>>
>> Erick
>>
>> ---------------------------------------------------------------------
>> To unsubscribe, e-mail: dev-unsubscribe@lucene.apache.org For additional
>> commands, e-mail: dev-help@lucene.apache.org
>
>
> ---------------------------------------------------------------------
> To unsubscribe, e-mail: dev-unsubscribe@lucene.apache.org
> For additional commands, e-mail: dev-help@lucene.apache.org
>
---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscribe@lucene.apache.org
For additional commands, e-mail: dev-help@lucene.apache.org
RE: Should we EOL support for 32 bit systems?
Posted by Uwe Schindler <uw...@thetaphi.de>.
Hi,
>From a Java point of view, there is no real difference to not support 32 bit versions. The bug with JDK 9 are just bugs that could also have happened with 64 bit versions. It is just easier to trigger this bug with 32 bits, but I am almost sure the underlying bug also affect 64 bits.
So why should we no longer support all platforms Java supports? Bitness does not matter for our code? Should we then also no longer support Sparc, PowerPC, or ARM platform?
-1 to add arbitrary restrictions on our runtime environment. If we want this, we should disallow all platforms we don't test on and of course also al processor variants we don't test on! :-)
Uwe
-----
Uwe Schindler
H.-H.-Meier-Allee 63, D-28213 Bremen
http://www.thetaphi.de
eMail: uwe@thetaphi.de
> -----Original Message-----
> From: Erick Erickson [mailto:erickerickson@gmail.com]
> Sent: Tuesday, August 25, 2015 6:23 PM
> To: dev@lucene.apache.org
> Subject: Should we EOL support for 32 bit systems?
>
> I have no real skin in this game, but I thought it worth asking after Uwe's
> recent e-mail about disabling 32bits with -server tests.
>
> I guess it boils down to "who is using 32-bit versions?". Are we spending
> time/energy supporting a configuration that is not useful to enough people
> to merit the effort?
>
> I'm perfectly content if the response is "That's a really stupid question to ask,
> of course we must continue to support 32-bit OSs".
> Although some evidence would be nice ;)
>
> It's just that nobody I work with is running 32 bit OS's. Whether that's just my
> limited exposure to people running small systems is certainly a valid
> question.
>
> If we _do_ decide to drop support for 32 bit systems, what's the right
> version? 6.0?
>
> Random thoughts on a slow Tuesday morning....
>
> Erick
>
> ---------------------------------------------------------------------
> To unsubscribe, e-mail: dev-unsubscribe@lucene.apache.org For additional
> commands, e-mail: dev-help@lucene.apache.org
---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscribe@lucene.apache.org
For additional commands, e-mail: dev-help@lucene.apache.org
Re: Should we EOL support for 32 bit systems?
Posted by Shawn Heisey <ap...@elyograg.org>.
On 8/25/2015 10:23 AM, Erick Erickson wrote:
> I have no real skin in this game, but I thought it worth asking after
> Uwe's recent e-mail about disabling 32bits with -server tests.
>
> I guess it boils down to "who is using 32-bit versions?". Are we
> spending time/energy supporting a configuration that is not useful to
> enough people to merit the effort?
>
> I'm perfectly content if the response is "That's a really stupid
> question to ask, of course we must continue to support 32-bit OSs".
> Although some evidence would be nice ;)
>
> It's just that nobody I work with is running 32 bit OS's. Whether
> that's just my limited exposure to people running small systems is
> certainly a valid question.
As long as Oracle has 32-bit versions of Java available for easy
download from the main website, we probably need to keep testing and
supporting it. Lucene and Solr won't scale on a 32-bit Java, but I
think there are still plenty of people who start there, and some who
actually run it on 32-bit.
That said, I think we do need to be thinking about dropping support in
the future, even though we can't really do so yet.
Intel stopped mass-producing 32-bit chips for the server market in 2005,
and stopped mass production of 32-bit chips for the consumer market in
2006. For nearly the last decade, it has been very difficult to buy a
computer incapable of 64-bit operation. Since Vista and Windows 7 came
on the scene, Microsoft has been pushing 64-bit client operating
systems. Server 2008R2 and Server 2012 are only available in 64-bit
editions. Macs have been 64-bit for a VERY long time.
I think the biggest market for 32-bit Java is browsers on Windows.
Virtually all installs of Firefox and Chrome for Windows are 32-bit, and
require the 32-bit Java. I bet that if the major browser vendors were
to all put out 64-bit versions on their main download links, downloads
of 32-bit Java would begin to dwindle rapidly.
On Windows 10, it looks like Microsoft has finally gone 64-bit by
default with the Edge browser. This might force the others to follow
suit. When that happens, I think Oracle may strongly consider dropping
32-bit support from the next major Java version ... and even if they
don't do it at that time, they probably will do so on the next major
version after that.
I just went to java.com with Microsoft Edge. It says "In Windows 10,
the Edge browser does not support plug-ins and therefore will not run
Java. Switch to a different browser (Firefox or Internet Explorer 11) to
run the Java plug-in." They are not going to be helpful in pushing
64-bit Java. :)
Thanks,
Shawn
---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscribe@lucene.apache.org
For additional commands, e-mail: dev-help@lucene.apache.org