RFR: JDK-8036003: Add --with-debug-symbols=[none|internal|external|zipped]

Wed Dec 9 00:31:19 UTC 2015

Hi,

I've uploaded a new webrev:
   http://cr.openjdk.java.net/~ysuenaga/JDK-8036003/webrev.06/

> Can you reinstate the comment:
>
>   581   # This must be done after the toolchain is setup, since we're looking at objcopy.
>
> as this constraint still exists.

I've fixed.

> I would have hoped that the deprecation macro enabled you to inform the user what the replacement is as per the comments! This seems unsatisfactory. :(

I added one more argument to BASIC_DEPRECATED_ARG_ENABLE to show message.
However, I'm not certain this change should be filed another issue or not.

Thanks,

Yasumasa

On 2015/12/03 15:51, David Holmes wrote:
> On 1/12/2015 12:34 AM, Yasumasa Suenaga wrote:
>> Hi Magnus,
>>
>> I've uploaded new webrev:
>>    http://cr.openjdk.java.net/~ysuenaga/JDK-8036003/webrev.05/
>
> common/autoconf/jdk-options.m4
>
> Can you reinstate the comment:
>
>   581   # This must be done after the toolchain is setup, since we're looking at objcopy.
>
> as this constraint still exists.
>
> It would be nice if the external and zipped cases could somehow be shared so that we don't have to duplicate everything just to alter the value of a single variable.
>
>   634   # --enable-zip-debug-info is deprecated.
>   635   # Please use --with-debug-symbols=zipped .
>   636   BASIC_DEPRECATED_ARG_ENABLE(zip-debug-info, zip_debug_info)
>
> I would have hoped that the deprecation macro enabled you to inform the user what the replacement is as per the comments! This seems unsatisfactory. :(
>
> I'm assuming there is no change to the default behaviour if none of these options are specified. I'm unclear whether we will need to coordinate this change with any internal processes that might explicitly be using the deprecated options - I think Magnus should have a better idea about that.
>
> Thanks,
> David
> -----
>
>>> If we do implement this flag, it should work as expected. If you
>>> believe this is too hard to get right, I'm alright with fixing this as
>>> a separate bug.
>>
>> Okay, I removed --enable-java-debug-symbols in this webrev.
>> I will file to JBS and try to create patch for it after this issue.
>>
>>
>>> That is, have you tried building with the four different values of
>>> --with-native-debug-symbols and verified that the result is indeed as
>>> you specified?
>>
>> Of course, I've checked all pattern of --with-native-debug-symbols.
>> I ran readelf -S for libjvm, libnio, libsaproc, and java command.
>> I checked debug section (.debug_info, and more) in ELF binaries.
>>
>>
>>> Just by my quick glance I can already spot what I believe is a
>>> mistake: for "none" you set STRIP_POLICY to min_strip, but it should
>>> reasonably be no_strip.
>>
>> I've fixed in this webrev.
>>
>>
>> Thanks,
>>
>> Yasumasa
>>
>>
>> On 2015/11/30 21:26, Magnus Ihse Bursie wrote:
>>>
>>>
>>> On 2015-11-26 15:19, Yasumasa Suenaga wrote:
>>>> I uploaded a new webrev:
>>>>   http://cr.openjdk.java.net/~ysuenaga/JDK-8036003/webrev.04/
>>>>
>>>> I implemented --with-native-debug-symbols and
>>>> --enable-java-debug-symbols in it.
>>>> And I deprecated --enable-debug-symbols and --enable-zip-debug-info .
>>>
>>> Hi Yasumasa,
>>>
>>> I apologize I have not been able to spend more time on this. I still
>>> only had time to take a short glimpse, but I found a few issues anyway.
>>>
>>> First, the new -enable-java-debug-symbols is problematic. If you build
>>> with "--enable-debug --disable-java-debug-symbols", then this implies
>>> that you would get a build without java debug symbols. However, this
>>> will not happen, since changing debug level wil enable java debug
>>> symbols, but the new java-debug-symbols flag will not manage to
>>> disable this.
>>>
>>> If we do implement this flag, it should work as expected. If you
>>> believe this is too hard to get right, I'm alright with fixing this as
>>> a separate bug.
>>>
>>> Second, all the flags we use to manage this internally makes me
>>> nervous. :) Are you sure you're getting them right? That is, have you
>>> tried building with the four different values of
>>> --with-native-debug-symbols and verified that the result is indeed as
>>> you specified? You'd need to check both libjvm.so and at least a jdk
>>> library, like libjava.so or so, since they use different build system
>>> and flags.
>>>
>>> Just by my quick glance I can already spot what I believe is a
>>> mistake: for "none" you set STRIP_POLICY to min_strip, but it should
>>> reasonably be no_strip. (I'd be surprised if strip was run if I
>>> compiled without debug symbols).
>>>
>>> Other than that, I believe this patch is starting to get ready.
>>>
>>> /Magnus
>>>>
>>>>
>>>> Thanks,
>>>>
>>>> Yasumasa
>>>>
>>>>
>>>> On 2015/11/25 23:52, Yasumasa Suenaga wrote:
>>>>> Hi Magnus,
>>>>>
>>>>>> --with-java-debug-symbols=none,all
>>>>>> (or "none,internal" for consistency).
>>>>>
>>>>> Can I change this option to --enable-java-debug-symbols ?
>>>>> This value set to none or all. So I think that we should use
>>>>> AC_ARG_ENABLE for it.
>>>>>
>>>>>
>>>>> Thanks,
>>>>>
>>>>> Yasumasa
>>>>>
>>>>>
>>>>> On 2015/11/25 23:35, Magnus Ihse Bursie wrote:
>>>>>> On 2015-11-21 14:12, Yasumasa Suenaga wrote:
>>>>>>>> My point with current proposal is that either it doesn't touch
>>>>>>>> the zipping because we already have an option for that; or it
>>>>>>>> should deprecate the existing option (now Erik has pointed out
>>>>>>>> that capability).
>>>>>>>
>>>>>>> I'm using BASIC_DEPRECATED_ARG_ENABLE for debug-symbols and
>>>>>>> zip-debug-info in new webrev:
>>>>>>> http://cr.openjdk.java.net/~ysuenaga/JDK-8036003/webrev.03/
>>>>>>>
>>>>>>> I had thought this new option as a wrapper for existing options.
>>>>>>> However, if current options should be deprecated, I will follow it.
>>>>>>
>>>>>> Hi Yasumasa,
>>>>>>
>>>>>> Sorry for the delay in my response.
>>>>>>
>>>>>> I had to go back and re-read the old discussions and my personal
>>>>>> notes about this. It turned out that some of the complications that
>>>>>> existed then is not relevant from Oracle's point of view anymore
>>>>>> (more specifically, stripping to several distinct levels). With
>>>>>> that complication gone, the basic case boils down to the three ways
>>>>>> of doing this:
>>>>>> 1) Compile without debug information ("none")
>>>>>> 2) Compile with debug information and keep it ("internal")
>>>>>> 3) Compile with debug information, copy it to an external file, and
>>>>>> strip the binary ("external")
>>>>>>
>>>>>> So far, so good. I agree with David Holmes that the zipping
>>>>>> complicates things somewhat. The problem here is two-fold: Firstly,
>>>>>> I believe that the "external" choice will be relevant to nobody (as
>>>>>> in, neither the community nor Oracle), and secondly, there have
>>>>>> been discussions of handling the external debuginfo differently
>>>>>> (namely, to move all debug info, unzipped, into a separate image,
>>>>>> and zip them all at once). It's a bit unfortunate if we create an
>>>>>> interface to configure that cannot grow properly with future changes.
>>>>>>
>>>>>> On the other hand, it does seem convenient to express the end
>>>>>> result for the Oracle build as "zipped" (or possibly
>>>>>> "external-zipped", to emphasize  the connection to "external"?).
>>>>>>
>>>>>> So I'm not sure what to think myself. But if David agrees that
>>>>>> adding "zipped" (or "external-zipped") to the new argument, and
>>>>>> deprecating the old, then I agree with it as well. :-)
>>>>>>
>>>>>> The second discussion here is the connection between java and
>>>>>> native debug symbols.
>>>>>>
>>>>>> Fact: We do not currently have a story for setting java debug
>>>>>> symbols. That's bad, and a regression as pointed out.
>>>>>> Fact: That does not mean it has to be fixed in this patch.
>>>>>>
>>>>>> I do think that java and native debug symbols are different things
>>>>>> and should be treated differently. But I also think we should have
>>>>>> configure arguments that show their similarity. So, I propose that
>>>>>> we have a
>>>>>> --with-native-debug-symbols=none,internal,external,zipped
>>>>>> as you propose, and then also a
>>>>>> --with-java-debug-symbols=none,all
>>>>>> (or "none,internal" for consistency).
>>>>>>
>>>>>> I think it would be nice if that flag could be implemented in the
>>>>>> same patch, but since it is a different thing I understand if you
>>>>>> want to leave it out. There is some added complication here, since
>>>>>> we already set -g for java compilation depending on debug level,
>>>>>> and the native debug symbol patch is obviously tricky as it is.
>>>>>>
>>>>>> /Magnus
>>>