Java bootstrap cleanups

Java bootstrap cleanups

[Gábor Boskovits]

[-- Attachment #1: Type: text/plain, Size: 1038 bytes --]

Hello guix,

I have a few questions/suggestions but I would like to get some feedback
before implementing them. This changeset would rebuild all java, so it
would be nice to get this right for the first time.

1. https://issues.guix.info/issue/30730 has a partial fix, icedtea-8 out
output is fixed. The same fix could be attempted at icedtea-7, and then we
could inherit it.
Also we could get rid of the repeated code in ant-bootstrap and co.
'strip-jar-timestamps the same way Ricardo did this in icedtea-8.
2. ISTM that https://issues.guix.info/issue/31719 is caused by not running
strip on the jre directory. I don't know how widespread this kind of issue
is throughout the codebase. If it is, then we might need to come up with a
more generic solution, but in the meanwhile I would simply add a
'strip-jre-directory phase, or something like that. I guess that should
interact with #: strip-binaries though.
3. some javadoc reproducibility issues could be mitigated by tweaking ant
build system to pass -notimestamp to javadoc.
WDYT?

[-- Attachment #2: Type: text/html, Size: 1321 bytes --]

Re: Java bootstrap cleanups

[Julien Lepiller]

Le 2018-09-11 09:32, Gábor Boskovits a écrit :
> Hello guix,
> 
> I have a few questions/suggestions but I would like to get some
> feedback before implementing them. This changeset would rebuild all
> java, so it would be nice to get this right for the first time.
> 
> 1. https://issues.guix.info/issue/30730 has a partial fix, icedtea-8
> out output is fixed. The same fix could be attempted at icedtea-7, and
> then we could inherit it.
> Also we could get rid of the repeated code in ant-bootstrap and co.
> 'strip-jar-timestamps the same way Ricardo did this in icedtea-8.
> 2. ISTM that https://issues.guix.info/issue/31719 is caused by not
> running strip on the jre directory. I don't know how widespread this
> kind of issue is throughout the codebase. If it is, then we might need
> to come up with a more generic solution, but in the meanwhile I would
> simply add a 'strip-jre-directory phase, or something like that. I
> guess that should interact with #: strip-binaries though.
> 3. some javadoc reproducibility issues could be mitigated by tweaking
> ant build system to pass -notimestamp to javadoc.
> WDYT?

I'd say go for it if that can fix the reproducibility issues we have :)

I don't expect 2. to be very widespread, but it happens elsewhere.
That's because some packages install private shared libraries (that's
weird, eh?). For instance, we have the same issue with ocaml-zarith:
`lib/ocaml/site-lib/zarith/dllzarith.so` is not stripped.

Re: Java bootstrap cleanups

[Ricardo Wurmus]

Hi Gábor,

> I have a few questions/suggestions but I would like to get some feedback
> before implementing them. This changeset would rebuild all java, so it
> would be nice to get this right for the first time.
>
> 1. https://issues.guix.info/issue/30730 has a partial fix, icedtea-8 out
> output is fixed. The same fix could be attempted at icedtea-7, and then we
> could inherit it.
> Also we could get rid of the repeated code in ant-bootstrap and co.
> 'strip-jar-timestamps the same way Ricardo did this in icedtea-8.

Yes, this seems like a good idea.

> 2. ISTM that https://issues.guix.info/issue/31719 is caused by not running
> strip on the jre directory. I don't know how widespread this kind of issue
> is throughout the codebase. If it is, then we might need to come up with a
> more generic solution, but in the meanwhile I would simply add a
> 'strip-jre-directory phase, or something like that. I guess that should
> interact with #: strip-binaries though.

Do you mean that the shared binaries that end up in the JRE directory
are not stripped and that stripping them gets rid of all these
references?  (You could test this by copying the binaries from the store
and stripping them manually.)

The icedtea packages use the gnu-build-system, so I would expect them to
be stripped by default.

> 3. some javadoc reproducibility issues could be mitigated by tweaking ant
> build system to pass -notimestamp to javadoc.
> WDYT?

Sounds good to me.

Thanks for working on this, Gábor!

--
Ricardo