Old Bug

[Jiri Vanek]

On 08/04/2012 10:23 AM, Thomas Meyer wrote:
> Am Freitag, den 03.08.2012, 15:23 +0100 schrieb Andrew Haley:
>> On 08/03/2012 03:10 PM, Jiri Vanek wrote:
>>> On 08/03/2012 02:13 PM, Jiri Vanek wrote:
>>>> On 08/03/2012 12:58 PM, Mark Wielaard wrote:
>>>>> On Fri, 2012-08-03 at 11:02 +0100, Andrew Haley wrote:
>>>>>> This is nothing to do with debuginfo.  It's to do with exported symbols.
>>>>>>
>>>>>> nm -D /usr/lib64/IcedTeaPlugin.so
>>>>>>
>>>>>> IcedTeaPlugin.so should only export the necessary symbols.
>>>>>
>>>>> This might be helpful:
>>>>> http://www.flameeyes.eu/autotools-mythbuster/libtool/symbols.html
>>>>>
>>>>> And of course lots on symbol visibility is described in:
>>>>> http://www.akkadia.org/drepper/dsohowto.pdf
>>>>
>>>> Yy, I found the same, good learning material
>>>>>
>>>>
>>>> So I'm Working on this.
>>>
>>> This is an attempt to remove unluckily exported symbols from icedtea-web plugin.
>>> However I do not know which symbols can be hidden and which can not. So this patch is removing the
>>> two originally reported.
>>>
>>> If to much symbols will have to be marked or to much should remains unmarked, then I will probably
>>> use blacklist/whitelist for this  de-exporting.
>>>
>>> this should go to 1.3 as well.
>>
>> We'll have to take all of them out that aren't explicitly required,
>> otherwise this bug will bite us again.
>>
>> Andrew.
>>
>
> attached patch should do the trick.
>
> with kind regards
> thomas
>
>
Hey! You have stolen may patch! :o) (and is nearly the same as I have prepared! O:)

It looks ok. The amount of symbols is very reduced..

I have just one  questions and one nitpick:


When i was searching for symbols which should remain visible - 
https://developer.mozilla.org/en-US/docs/Gecko_Plugin_API_Reference, 
https://developer.mozilla.org/en-US/docs/Gecko_Plugin_API_Reference/Scripting_plugins it looked to 
me that more should remain marked as "default" when fvisibility=hidden is used.  You are marking 
just three of them. Are you sure with them to be enough?
Anyway -  applets and regression tests  looks working fine, so I'm ok for head.

nitpisck - do you think it is worthy to unify old and new C switches to new variable?
eg PLUGIN_COMPILATION_SWITCHES= -fvisibility=hidden -fPIC

Best regards,
   J.