Default value for -with_native_debug_symbols on Windows useful?

classic Classic list List threaded Threaded
9 messages Options
Reply | Threaded
Open this post in threaded view
|

Default value for -with_native_debug_symbols on Windows useful?

Thomas Stüfe-2
Hi all,

I could reframe the subject matter as "how do you internally at Oracle
build the Windows JDK"?

I had several discussions with people from Oracle about missing callstacks
in hs-err files on Windows generated in tests which run at Oracle.

I recently did JDK-8185712, which improves the Windows symbol decoder, so I
feel somewhat responsible. However, I was never able to reproduce anything
- here at SAP, we get nice callstacks and all works well.

A simple explanation may be that you at Oracle either remove the debug info
(pdb files) from the images before testing and/or build
with -with_native_debug_symbols=zipped . The latter zips the pdb files,
which in that form are useless. Which is it?

For that matter, do you think -with_native_debug_symbols=zipped makes sense
on Windows? Would it not make more sense to make the default
-with_native_debug_symbols=external on Windows, to have pdb files ready for
debugging and testing?

Thanks, Thomas
Reply | Threaded
Open this post in threaded view
|

Re: Default value for -with_native_debug_symbols on Windows useful?

Erik Joelsson
I agree that the zipped debug symbols are just annoying and should be
removed.

/Erik


On 2017-11-08 22:32, Thomas Stüfe wrote:

> Hi all,
>
> I could reframe the subject matter as "how do you internally at Oracle
> build the Windows JDK"?
>
> I had several discussions with people from Oracle about missing callstacks
> in hs-err files on Windows generated in tests which run at Oracle.
>
> I recently did JDK-8185712, which improves the Windows symbol decoder, so I
> feel somewhat responsible. However, I was never able to reproduce anything
> - here at SAP, we get nice callstacks and all works well.
>
> A simple explanation may be that you at Oracle either remove the debug info
> (pdb files) from the images before testing and/or build
> with -with_native_debug_symbols=zipped . The latter zips the pdb files,
> which in that form are useless. Which is it?
>
> For that matter, do you think -with_native_debug_symbols=zipped makes sense
> on Windows? Would it not make more sense to make the default
> -with_native_debug_symbols=external on Windows, to have pdb files ready for
> debugging and testing?
>
> Thanks, Thomas

Reply | Threaded
Open this post in threaded view
|

Re: Default value for -with_native_debug_symbols on Windows useful?

Thomas Stüfe-2
Do you think this for all platforms or just for Windows? With removed, do
you really mean the feature removed, not just the default changed?

Thomas

On Thu 9. Nov 2017 at 19:58, Erik Joelsson <[hidden email]> wrote:

> I agree that the zipped debug symbols are just annoying and should be
> removed.
>
> /Erik
>
>
> On 2017-11-08 22:32, Thomas Stüfe wrote:
> > Hi all,
> >
> > I could reframe the subject matter as "how do you internally at Oracle
> > build the Windows JDK"?
> >
> > I had several discussions with people from Oracle about missing
> callstacks
> > in hs-err files on Windows generated in tests which run at Oracle.
> >
> > I recently did JDK-8185712, which improves the Windows symbol decoder,
> so I
> > feel somewhat responsible. However, I was never able to reproduce
> anything
> > - here at SAP, we get nice callstacks and all works well.
> >
> > A simple explanation may be that you at Oracle either remove the debug
> info
> > (pdb files) from the images before testing and/or build
> > with -with_native_debug_symbols=zipped . The latter zips the pdb files,
> > which in that form are useless. Which is it?
> >
> > For that matter, do you think -with_native_debug_symbols=zipped makes
> sense
> > on Windows? Would it not make more sense to make the default
> > -with_native_debug_symbols=external on Windows, to have pdb files ready
> for
> > debugging and testing?
> >
> > Thanks, Thomas
>
>
Reply | Threaded
Open this post in threaded view
|

Re: Default value for -with_native_debug_symbols on Windows useful?

Erik Joelsson
Hello,


On 2017-11-09 15:26, Thomas Stüfe wrote:
> Do you think this for all platforms or just for Windows? With removed,
> do you really mean the feature removed, not just the default changed?
>
I mean for all platforms. I don't see any point in the build wasting
time on zipping up the debug symbols which also makes them unusable. If
someone cries out we could leave the feature in, but I would much rather
not. Note that the bundle target has to unzip them again to put them in
the symbols bundle, which is the final deliverable of the symbols.

Historically there have been various reasons for this zipping internally
to Oracle, but I don't think any of those reasons hold up anymore in our
current JDK 10 infrastructure.

/Erik

> Thomas
>
> On Thu 9. Nov 2017 at 19:58, Erik Joelsson <[hidden email]
> <mailto:[hidden email]>> wrote:
>
>     I agree that the zipped debug symbols are just annoying and should be
>     removed.
>
>     /Erik
>
>
>     On 2017-11-08 22:32, Thomas Stüfe wrote:
>     > Hi all,
>     >
>     > I could reframe the subject matter as "how do you internally at
>     Oracle
>     > build the Windows JDK"?
>     >
>     > I had several discussions with people from Oracle about missing
>     callstacks
>     > in hs-err files on Windows generated in tests which run at Oracle.
>     >
>     > I recently did JDK-8185712, which improves the Windows symbol
>     decoder, so I
>     > feel somewhat responsible. However, I was never able to
>     reproduce anything
>     > - here at SAP, we get nice callstacks and all works well.
>     >
>     > A simple explanation may be that you at Oracle either remove the
>     debug info
>     > (pdb files) from the images before testing and/or build
>     > with -with_native_debug_symbols=zipped . The latter zips the pdb
>     files,
>     > which in that form are useless. Which is it?
>     >
>     > For that matter, do you think -with_native_debug_symbols=zipped
>     makes sense
>     > on Windows? Would it not make more sense to make the default
>     > -with_native_debug_symbols=external on Windows, to have pdb
>     files ready for
>     > debugging and testing?
>     >
>     > Thanks, Thomas
>

Reply | Threaded
Open this post in threaded view
|

Re: Default value for -with_native_debug_symbols on Windows useful?

Thomas Stüfe-2
Hi Erik,

On Fri, Nov 10, 2017 at 12:52 AM, Erik Joelsson <[hidden email]>
wrote:

> Hello,
>
> On 2017-11-09 15:26, Thomas Stüfe wrote:
>
> Do you think this for all platforms or just for Windows? With removed, do
> you really mean the feature removed, not just the default changed?
>
> I mean for all platforms. I don't see any point in the build wasting time
> on zipping up the debug symbols which also makes them unusable. If someone
> cries out we could leave the feature in, but I would much rather not. Note
> that the bundle target has to unzip them again to put them in the symbols
> bundle, which is the final deliverable of the symbols.
>
> Historically there have been various reasons for this zipping internally
> to Oracle, but I don't think any of those reasons hold up anymore in our
> current JDK 10 infrastructure.
>

I am all for it. We never really used that feature.

Best Regards, Thomas


>
>
> /Erik
>
> Thomas
>
> On Thu 9. Nov 2017 at 19:58, Erik Joelsson <[hidden email]>
> wrote:
>
>> I agree that the zipped debug symbols are just annoying and should be
>> removed.
>>
>> /Erik
>>
>>
>> On 2017-11-08 22:32, Thomas Stüfe wrote:
>> > Hi all,
>> >
>> > I could reframe the subject matter as "how do you internally at Oracle
>> > build the Windows JDK"?
>> >
>> > I had several discussions with people from Oracle about missing
>> callstacks
>> > in hs-err files on Windows generated in tests which run at Oracle.
>> >
>> > I recently did JDK-8185712, which improves the Windows symbol decoder,
>> so I
>> > feel somewhat responsible. However, I was never able to reproduce
>> anything
>> > - here at SAP, we get nice callstacks and all works well.
>> >
>> > A simple explanation may be that you at Oracle either remove the debug
>> info
>> > (pdb files) from the images before testing and/or build
>> > with -with_native_debug_symbols=zipped . The latter zips the pdb files,
>> > which in that form are useless. Which is it?
>> >
>> > For that matter, do you think -with_native_debug_symbols=zipped makes
>> sense
>> > on Windows? Would it not make more sense to make the default
>> > -with_native_debug_symbols=external on Windows, to have pdb files
>> ready for
>> > debugging and testing?
>> >
>> > Thanks, Thomas
>>
>>
>
Reply | Threaded
Open this post in threaded view
|

Re: Default value for -with_native_debug_symbols on Windows useful?

Erik Helin-2
On 11/10/2017 09:01 AM, Thomas Stüfe wrote:

> Hi Erik,
>
> On Fri, Nov 10, 2017 at 12:52 AM, Erik Joelsson <[hidden email]>
> wrote:
>
>> Hello,
>>
>> On 2017-11-09 15:26, Thomas Stüfe wrote:
>>
>> Do you think this for all platforms or just for Windows? With removed, do
>> you really mean the feature removed, not just the default changed?
>>
>> I mean for all platforms. I don't see any point in the build wasting time
>> on zipping up the debug symbols which also makes them unusable. If someone
>> cries out we could leave the feature in, but I would much rather not. Note
>> that the bundle target has to unzip them again to put them in the symbols
>> bundle, which is the final deliverable of the symbols.
>>
>> Historically there have been various reasons for this zipping internally
>> to Oracle, but I don't think any of those reasons hold up anymore in our
>> current JDK 10 infrastructure.
>>
>
> I am all for it. We never really used that feature.

Could we start by changing the default (for all platforms) to be either
'external' or 'internal' and then work on eventually removing 'zipped'?

Thanks,
Erik

> Best Regards, Thomas
>
>
>>
>>
>> /Erik
>>
>> Thomas
>>
>> On Thu 9. Nov 2017 at 19:58, Erik Joelsson <[hidden email]>
>> wrote:
>>
>>> I agree that the zipped debug symbols are just annoying and should be
>>> removed.
>>>
>>> /Erik
>>>
>>>
>>> On 2017-11-08 22:32, Thomas Stüfe wrote:
>>>> Hi all,
>>>>
>>>> I could reframe the subject matter as "how do you internally at Oracle
>>>> build the Windows JDK"?
>>>>
>>>> I had several discussions with people from Oracle about missing
>>> callstacks
>>>> in hs-err files on Windows generated in tests which run at Oracle.
>>>>
>>>> I recently did JDK-8185712, which improves the Windows symbol decoder,
>>> so I
>>>> feel somewhat responsible. However, I was never able to reproduce
>>> anything
>>>> - here at SAP, we get nice callstacks and all works well.
>>>>
>>>> A simple explanation may be that you at Oracle either remove the debug
>>> info
>>>> (pdb files) from the images before testing and/or build
>>>> with -with_native_debug_symbols=zipped . The latter zips the pdb files,
>>>> which in that form are useless. Which is it?
>>>>
>>>> For that matter, do you think -with_native_debug_symbols=zipped makes
>>> sense
>>>> on Windows? Would it not make more sense to make the default
>>>> -with_native_debug_symbols=external on Windows, to have pdb files
>>> ready for
>>>> debugging and testing?
>>>>
>>>> Thanks, Thomas
>>>
>>>
>>
Reply | Threaded
Open this post in threaded view
|

Re: Default value for -with_native_debug_symbols on Windows useful?

Thomas Stüfe-2
On Fri, Nov 10, 2017 at 3:50 PM, Erik Helin <[hidden email]> wrote:

> On 11/10/2017 09:01 AM, Thomas Stüfe wrote:
>
>> Hi Erik,
>>
>> On Fri, Nov 10, 2017 at 12:52 AM, Erik Joelsson <[hidden email]
>> >
>> wrote:
>>
>> Hello,
>>>
>>> On 2017-11-09 15:26, Thomas Stüfe wrote:
>>>
>>> Do you think this for all platforms or just for Windows? With removed, do
>>> you really mean the feature removed, not just the default changed?
>>>
>>> I mean for all platforms. I don't see any point in the build wasting time
>>> on zipping up the debug symbols which also makes them unusable. If
>>> someone
>>> cries out we could leave the feature in, but I would much rather not.
>>> Note
>>> that the bundle target has to unzip them again to put them in the symbols
>>> bundle, which is the final deliverable of the symbols.
>>>
>>> Historically there have been various reasons for this zipping internally
>>> to Oracle, but I don't think any of those reasons hold up anymore in our
>>> current JDK 10 infrastructure.
>>>
>>>
>> I am all for it. We never really used that feature.
>>
>
> Could we start by changing the default (for all platforms) to be either
> 'external' or 'internal' and then work on eventually removing 'zipped'?
>
>
Sure. Gives us an easy way back should someone complain :)

Cheers, Thomas


> Thanks,
> Erik
>
>
> Best Regards, Thomas
>>
>>
>>
>>>
>>> /Erik
>>>
>>> Thomas
>>>
>>> On Thu 9. Nov 2017 at 19:58, Erik Joelsson <[hidden email]>
>>> wrote:
>>>
>>> I agree that the zipped debug symbols are just annoying and should be
>>>> removed.
>>>>
>>>> /Erik
>>>>
>>>>
>>>> On 2017-11-08 22:32, Thomas Stüfe wrote:
>>>>
>>>>> Hi all,
>>>>>
>>>>> I could reframe the subject matter as "how do you internally at Oracle
>>>>> build the Windows JDK"?
>>>>>
>>>>> I had several discussions with people from Oracle about missing
>>>>>
>>>> callstacks
>>>>
>>>>> in hs-err files on Windows generated in tests which run at Oracle.
>>>>>
>>>>> I recently did JDK-8185712, which improves the Windows symbol decoder,
>>>>>
>>>> so I
>>>>
>>>>> feel somewhat responsible. However, I was never able to reproduce
>>>>>
>>>> anything
>>>>
>>>>> - here at SAP, we get nice callstacks and all works well.
>>>>>
>>>>> A simple explanation may be that you at Oracle either remove the debug
>>>>>
>>>> info
>>>>
>>>>> (pdb files) from the images before testing and/or build
>>>>> with -with_native_debug_symbols=zipped . The latter zips the pdb
>>>>> files,
>>>>> which in that form are useless. Which is it?
>>>>>
>>>>> For that matter, do you think -with_native_debug_symbols=zipped makes
>>>>>
>>>> sense
>>>>
>>>>> on Windows? Would it not make more sense to make the default
>>>>> -with_native_debug_symbols=external on Windows, to have pdb files
>>>>>
>>>> ready for
>>>>
>>>>> debugging and testing?
>>>>>
>>>>> Thanks, Thomas
>>>>>
>>>>
>>>>
>>>>
>>>
Reply | Threaded
Open this post in threaded view
|

Re: Default value for -with_native_debug_symbols on Windows useful?

Magnus Ihse Bursie
In reply to this post by Erik Helin-2

On 2017-11-10 15:50, Erik Helin wrote:

> On 11/10/2017 09:01 AM, Thomas Stüfe wrote:
>> Hi Erik,
>>
>> On Fri, Nov 10, 2017 at 12:52 AM, Erik Joelsson
>> <[hidden email]>
>> wrote:
>>
>>> Hello,
>>>
>>> On 2017-11-09 15:26, Thomas Stüfe wrote:
>>>
>>> Do you think this for all platforms or just for Windows? With
>>> removed, do
>>> you really mean the feature removed, not just the default changed?
>>>
>>> I mean for all platforms. I don't see any point in the build wasting
>>> time
>>> on zipping up the debug symbols which also makes them unusable. If
>>> someone
>>> cries out we could leave the feature in, but I would much rather
>>> not. Note
>>> that the bundle target has to unzip them again to put them in the
>>> symbols
>>> bundle, which is the final deliverable of the symbols.
>>>
>>> Historically there have been various reasons for this zipping
>>> internally
>>> to Oracle, but I don't think any of those reasons hold up anymore in
>>> our
>>> current JDK 10 infrastructure.
>>>
>>
>> I am all for it. We never really used that feature.
>
> Could we start by changing the default (for all platforms) to be
> either 'external' or 'internal' and then work on eventually removing
> 'zipped'?

I think that's the best way forward. I created
https://bugs.openjdk.java.net/browse/JDK-8191205. I'd like to keep the
default to "external", to have that code excercised (since it's more
complex than "internal").

/Magnus

>
> Thanks,
> Erik
>
>> Best Regards, Thomas
>>
>>
>>>
>>>
>>> /Erik
>>>
>>> Thomas
>>>
>>> On Thu 9. Nov 2017 at 19:58, Erik Joelsson <[hidden email]>
>>> wrote:
>>>
>>>> I agree that the zipped debug symbols are just annoying and should be
>>>> removed.
>>>>
>>>> /Erik
>>>>
>>>>
>>>> On 2017-11-08 22:32, Thomas Stüfe wrote:
>>>>> Hi all,
>>>>>
>>>>> I could reframe the subject matter as "how do you internally at
>>>>> Oracle
>>>>> build the Windows JDK"?
>>>>>
>>>>> I had several discussions with people from Oracle about missing
>>>> callstacks
>>>>> in hs-err files on Windows generated in tests which run at Oracle.
>>>>>
>>>>> I recently did JDK-8185712, which improves the Windows symbol
>>>>> decoder,
>>>> so I
>>>>> feel somewhat responsible. However, I was never able to reproduce
>>>> anything
>>>>> - here at SAP, we get nice callstacks and all works well.
>>>>>
>>>>> A simple explanation may be that you at Oracle either remove the
>>>>> debug
>>>> info
>>>>> (pdb files) from the images before testing and/or build
>>>>> with -with_native_debug_symbols=zipped . The latter zips the pdb
>>>>> files,
>>>>> which in that form are useless. Which is it?
>>>>>
>>>>> For that matter, do you think -with_native_debug_symbols=zipped makes
>>>> sense
>>>>> on Windows? Would it not make more sense to make the default
>>>>> -with_native_debug_symbols=external on Windows, to have pdb files
>>>> ready for
>>>>> debugging and testing?
>>>>>
>>>>> Thanks, Thomas
>>>>
>>>>
>>>

Reply | Threaded
Open this post in threaded view
|

Re: Default value for -with_native_debug_symbols on Windows useful?

Thomas Stüfe-2
Thanks Markus!

On Tue, Nov 14, 2017 at 1:30 PM, Magnus Ihse Bursie <
[hidden email]> wrote:

>
> On 2017-11-10 15:50, Erik Helin wrote:
>
>> On 11/10/2017 09:01 AM, Thomas Stüfe wrote:
>>
>>> Hi Erik,
>>>
>>> On Fri, Nov 10, 2017 at 12:52 AM, Erik Joelsson <
>>> [hidden email]>
>>> wrote:
>>>
>>> Hello,
>>>>
>>>> On 2017-11-09 15:26, Thomas Stüfe wrote:
>>>>
>>>> Do you think this for all platforms or just for Windows? With removed,
>>>> do
>>>> you really mean the feature removed, not just the default changed?
>>>>
>>>> I mean for all platforms. I don't see any point in the build wasting
>>>> time
>>>> on zipping up the debug symbols which also makes them unusable. If
>>>> someone
>>>> cries out we could leave the feature in, but I would much rather not.
>>>> Note
>>>> that the bundle target has to unzip them again to put them in the
>>>> symbols
>>>> bundle, which is the final deliverable of the symbols.
>>>>
>>>> Historically there have been various reasons for this zipping internally
>>>> to Oracle, but I don't think any of those reasons hold up anymore in our
>>>> current JDK 10 infrastructure.
>>>>
>>>>
>>> I am all for it. We never really used that feature.
>>>
>>
>> Could we start by changing the default (for all platforms) to be either
>> 'external' or 'internal' and then work on eventually removing 'zipped'?
>>
>
> I think that's the best way forward. I created
> https://bugs.openjdk.java.net/browse/JDK-8191205. I'd like to keep the
> default to "external", to have that code excercised (since it's more
> complex than "internal").
>
> /Magnus
>
>
>
>> Thanks,
>> Erik
>>
>> Best Regards, Thomas
>>>
>>>
>>>
>>>>
>>>> /Erik
>>>>
>>>> Thomas
>>>>
>>>> On Thu 9. Nov 2017 at 19:58, Erik Joelsson <[hidden email]>
>>>> wrote:
>>>>
>>>> I agree that the zipped debug symbols are just annoying and should be
>>>>> removed.
>>>>>
>>>>> /Erik
>>>>>
>>>>>
>>>>> On 2017-11-08 22:32, Thomas Stüfe wrote:
>>>>>
>>>>>> Hi all,
>>>>>>
>>>>>> I could reframe the subject matter as "how do you internally at Oracle
>>>>>> build the Windows JDK"?
>>>>>>
>>>>>> I had several discussions with people from Oracle about missing
>>>>>>
>>>>> callstacks
>>>>>
>>>>>> in hs-err files on Windows generated in tests which run at Oracle.
>>>>>>
>>>>>> I recently did JDK-8185712, which improves the Windows symbol decoder,
>>>>>>
>>>>> so I
>>>>>
>>>>>> feel somewhat responsible. However, I was never able to reproduce
>>>>>>
>>>>> anything
>>>>>
>>>>>> - here at SAP, we get nice callstacks and all works well.
>>>>>>
>>>>>> A simple explanation may be that you at Oracle either remove the debug
>>>>>>
>>>>> info
>>>>>
>>>>>> (pdb files) from the images before testing and/or build
>>>>>> with -with_native_debug_symbols=zipped . The latter zips the pdb
>>>>>> files,
>>>>>> which in that form are useless. Which is it?
>>>>>>
>>>>>> For that matter, do you think -with_native_debug_symbols=zipped makes
>>>>>>
>>>>> sense
>>>>>
>>>>>> on Windows? Would it not make more sense to make the default
>>>>>> -with_native_debug_symbols=external on Windows, to have pdb files
>>>>>>
>>>>> ready for
>>>>>
>>>>>> debugging and testing?
>>>>>>
>>>>>> Thanks, Thomas
>>>>>>
>>>>>
>>>>>
>>>>>
>>>>
>