Windows build failure in JDK8 with --disable-zip-debug-info

[Volker Simonis]

Hi Dmitry,

no, they are not the same. I forgot to mention that.
On MinGW we have 3.81 and on Cygwin 4.1

Are there any known problems with the new Version?

I specially use GNU make 4.1 on Cygwin because it can nicely handle
":" in target path names (unfortunately we still have them in our
build :(. With older version of make this leads to "multiple target
patterns" errors.

Regards,
Volker


On Thu, Mar 12, 2015 at 3:39 PM, Dmitry Samersoff
<dmitry.samersoff at oracle.com> wrote:
> Volker,
>
> What version of *make* do you use in both cases?
>
> -Dmitry
>
> On 2015-03-12 17:34, Volker Simonis wrote:
>> Hi,
>>
>> I understand that I'm a little late to the game but I just run into
>> this problem myself:)
>>
>> The funny thing is that this problem doesn't occur with MinGW/MSYS but
>> just with Cygwin and I can't understand why?
>>
>> We have a little special setup here at SAP: we do the Windows builds
>> with MinGW/MSYS and by default we always build with
>> --disable-zip-debug-info. So until now we had no problems.
>>
>> Now I started to migrate our build to Cygwin (but still with
>> --disable-zip-debug-info) and run into the problem.
>>
>> I think the origin of the dependency on the .map (and .pdb) files is
>> clear - it is set right in SetupNativeCompilation if we don't want to
>> zip the debug information (i.e. --disable-zip-debug-info):
>>
>>         ifeq ($(ZIP_DEBUGINFO_FILES), true)
>>           ...
>>         else
>>           ifeq ($(OPENJDK_TARGET_OS), windows)
>>             $1 += $$($1_OUTPUT_DIR)/$$($1_LIBRARY).map \
>>                 $$($1_OUTPUT_DIR)/$$($1_LIBRARY).pdb
>>
>> But we also have the following pattern rule in SetupNativeCompilation
>> which should copy the .map and .pdb from OBJECT_DIR to OUTPUT_DIR:
>>
>>         ifneq ($$($1_OUTPUT_DIR),$$($1_OBJECT_DIR))
>>           $$($1_OUTPUT_DIR)/% : $$($1_OBJECT_DIR)/%
>>           $(CP) $$< $$@
>>         endif
>>
>> This rule works perfectly with MinGW/MSYS but it doesn't get triggered
>> with Cygwin. And that's the reason why we get the error "*** No rule
>> to make target `/cygdrive/c/jprt/T/P1/031627.daholme/s/build/windows-x86-normal-clientANDserver-release/jdk/bin/verify.map'"
>> which says that it can not make the .map file in the OUTPUT_DIR. But
>> notice that the .map file is there in the OBJECT_DIR directory (i.e.
>> ../objs/libverify/verify.map), but make under Cygwin somehow doesn't
>> recognize that there's a rule to copy it over to the OUTPUT_DIR
>> directory.
>>
>> I tried the workaround proposed by Magnus' but unfortunately it
>> doesn't work. I think that's because the problem is not that the .map
>> files are not created - the problem is that they are not copied over
>> to the OUTPUT_DIR.
>>
>> So here's what really helped:
>>
>>           ifeq ($(OPENJDK_TARGET_OS), windows)
>>             $1 += $$($1_OUTPUT_DIR)/$$($1_LIBRARY).map \
>>                 $$($1_OUTPUT_DIR)/$$($1_LIBRARY).pdb
>>
>>             $$($1_OUTPUT_DIR)/$$($1_LIBRARY).map :
>> $$($1_OBJECT_DIR)/$$($1_LIBRARY).map
>>               echo "Copying .map from OBJECT_DIR to OUTPUT_DIR"
>>               $(CP) $$< $$@
>>             $$($1_OUTPUT_DIR)/$$($1_LIBRARY).pdb :
>> $$($1_OBJECT_DIR)/$$($1_LIBRARY).pdb
>>                echo "Copying .pdb from OBJECT_DIR to OUTPUT_DIR"
>>                $(CP) $$< $$@
>>
>> And you need the same fix for the PROGRAM build part:
>>
>>           ifeq ($(OPENJDK_TARGET_OS), windows)
>>             $1 += $$($1_OUTPUT_DIR)/$$($1_PROGRAM).map \
>>                 $$($1_OUTPUT_DIR)/$$($1_PROGRAM).pdb
>>
>>             $$($1_OUTPUT_DIR)/$$($1_PROGRAM).map :
>> $$($1_OBJECT_DIR)/$$($1_PROGRAM).map
>>                echo "Copying .map from OBJECT_DIR to OUTPUT_DIR"
>>                $(CP) $$< $$@
>>             $$($1_OUTPUT_DIR)/$$($1_PROGRAM).pdb :
>> $$($1_OBJECT_DIR)/$$($1_PROGRAM).pdb
>>                echo "Copying .pdb from OBJECT_DIR to OUTPUT_DIR"
>>                $(CP) $$< $$@
>>
>> So for me this works now and I will change our internal build accordingly.
>>
>> I don't know if there's any interest of bringing this to jdk8u. I just
>> though I'll let you know:)
>>
>> It would also be interesting if somebody has some explanation for why
>> the pattern rule for copying the .map files works under MinGW/MSYS but
>> not under Cygwin.
>>
>> Regards,
>> Volker
>>
>> On Wed, Nov 5, 2014 at 2:00 PM, David Holmes <david.holmes at oracle.com> wrote:
>>> On 5/11/2014 10:27 PM, Magnus Ihse Bursie wrote:
>>>>
>>>> On 2014-11-05 13:25, Magnus Ihse Bursie wrote:
>>>>>
>>>>> I even have a vague memory of a fix along these lines in jdk9. If
>>>>> that's correct, it's probably due for backporting. I'll see if I can
>>>>> locate it.
>>>>
>>>>
>>>> https://bugs.openjdk.java.net/browse/JDK-8025936
>>>>
>>>> It might be some work backporting it though, the comments in the bug
>>>> says it needed to be substantially rewritten due to changes in JDK9.
>>>
>>>
>>> Thanks Magnus. Seems this is unlikely to be fixed then - which means I can't
>>> test what I'm working on for the case where we don't zip the debuginfo files
>>> :(
>>>
>>> David
>>>
>>>> /Magnus
>
>
> --
> Dmitry Samersoff
> Oracle Java development team, Saint Petersburg, Russia
> * I would love to change the world, but they won't give me the sources.