RFE/RFC reproducible policies
erik.joelsson at oracle.com
Fri May 20 08:12:50 UTC 2016
(Adding Brad from the team owning these files)
I think I understand the problem now. Each time we create the jars, they
come out with binary differences because there are different timestamps
on the file inside them, while the files inside them are actually the
same. Does the timestamp on the jar files matter or just the contents of
the jar files? I can't imagine jlink/jmod affecting the timestamps
inside the jars, only the jars themselves.
Regarding GendataPolicyJars.gmk vs CreateSecurityJars.gmk, one is in JDK
8, the other in JDK 9.
Regarding your proposed change, I think the security team needs to
approve the approach before I judge it from a build point of view.
On 2016-05-20 09:25, Jiri Vanek wrote:
> Thank you for quick answer.
> However. .not exactly the one I hoped for:(
> Generally - IMHO the CreateSecurityJars.gmk is never used to generate
> policies, only the GendataPolicyJars.gmk is used for that.
> Thats why my unhappy patch is here.
> In one way or another - the GendataPolicyJars should be removed, or he
> lines regarding policies' jars from CreateSecurityJars should e
> removed (second is probably more correct way, If you light on green
> light to this patch, I will remove the lines from CreateSecurityJars
> and test)
> jmod? jlink? Crap :) The policies had remained simple jars, and are
> nothing but copied into image. I was looking to Images.gmk pretty
> much, and found quite a usages of JLINK_TOOL but found much less
> looking to it,, not its usages nor it documentation did not resolved
> how the jars get to the old good.../lib/security/ directory. It does
> not have much to do with modules....
> And even if the jlink tool should be responsible for copying the jars
> (which is pretty obfuscated way) then perhaps it should keep time
> stamps? (just thought)
> I'm much much more troubled about way how the GendataPolicyJars and
> CreateSecurityJars are mixed up:(
> Again, thank you very much for looking into it!
> On 05/19/2016 07:07 PM, Erik Joelsson wrote:
>> Hello Jiri,
>> If I understand the question correctly, you are wondering how the
>> policy files from
>> CreateSecurityJars.gmk end up in the final image? This is done in two
>> steps. First the new JDK 9
>> tool jmod packages each module into a distribution format (typically
>> java.base.jmod). Then the next
>> new tool jlink links all the jmods together to create an image.
>> Somewhere inside those tools, I
>> assume timestamps are changing.
>> On 2016-05-19 18:52, Jiri Vanek wrote:
>>> Hello again!
>>> Recent Feature Complete milestone have scared me, as I have
>>> long-time persisting issue when
>>> packaging openjdk (6..7...8 and 9)
>>> The policy jars, always from same source, never the same. As they
>>> are considered as configure
>>> files, the RPM update treat them alike.
>>> Not so do jdk build system, and every build have its "special" but
>>> still the same. .policies.
>>> This is fixed in my RPMS since  like 
>>> Well, not nice. I checked icedtea, and they since  already have
>>> this change 
>>> So I looked into JDK9 and.. it ahave teh change in
>>> CreateSecurityJars.gmk ! Not whole, but
>>> definitely not used. I really do not understand why.
>>> So there is patch for jdk9's -
>>> which is making the
>>> policies truly static even with all this necessary stamping.
>>> However, I must apologise for missing part, which I had not found
>>> how to solve.
>>> Up to "make" (build) everything is ok. but "make images" corrupts
>>> the timestamps, I did not
>>> found, where the built files flow to images:(to stamp them again,
>>> and last time)
>>> Best regards from CZ
More information about the build-dev