Re: RFR: 8288001: SOURCE_DATE_EPOCH should not be set by default
On Mon, 13 Jun 2022 06:50:54 GMT, KIRIYAMA Takuya wrote: >> At default configuration, SOURCE_DATE_EPOCH is exported as environment >> variable in SetupReproducibleBuild. Then, gcc is affected of >> SOURCE_DATE_EPOCH environment variable. This value is used only to set >> SOURCE_DATE_ISO_8601 (except below), so I removed "export" for >> SOURCE_DATE_EPOCH in SetupReproducibleBuild. And, at building ct.sym, >> SOURCE_DATE_EPOCH environment variable is needed. So I added setting routine >> if SOURCE_DATE_EPOCH is not set. >> >> This fix works fine. With default configuration shows -Xinternalversion >> output same as Windows, and with --with-source-date configuration shows >> -Xinternalversion output specified timestamp. Would you please review this >> fix? > > Thank you for your comments. > I understood the goal of reproducible build. But now, > ENABLE_REPRODUCIBLE_BUILD is set to false in default configuration. > Then I think minimize the effort of SOURCE_DATE_EPOCH when reproducible build > is disabled. I wonder why build time information is different from Windows > and Linux. @tkiriyama I have created https://bugs.openjdk.org/browse/JDK-8288396 for the approach I championed above, i.e. to always build reproducible instead. As part of this, the `-Xinternalversion` time stamp should be of the same format for all platforms, given the same set of options to `configure`. The associated PR is https://github.com/openjdk/jdk/pull/9152. Can you verify if this solves your issues, and if so, close this bug? - PR: https://git.openjdk.org/jdk/pull/9081
Re: RFR: 8288001: SOURCE_DATE_EPOCH should not be set by default
On Wed, 8 Jun 2022 07:47:24 GMT, KIRIYAMA Takuya wrote: > At default configuration, SOURCE_DATE_EPOCH is exported as environment > variable in SetupReproducibleBuild. Then, gcc is affected of > SOURCE_DATE_EPOCH environment variable. This value is used only to set > SOURCE_DATE_ISO_8601 (except below), so I removed "export" for > SOURCE_DATE_EPOCH in SetupReproducibleBuild. And, at building ct.sym, > SOURCE_DATE_EPOCH environment variable is needed. So I added setting routine > if SOURCE_DATE_EPOCH is not set. > > This fix works fine. With default configuration shows -Xinternalversion > output same as Windows, and with --with-source-date configuration shows > -Xinternalversion output specified timestamp. Would you please review this > fix? Is the problem that on Windows we get -Xinternalversion in local timezone but on Linux we get UTC? That's a difference that may warrant fixing, but I don't think this is the way to do it. I don't think it makes a practical difference if the timestamp is taken at the start of the build or when a certain src file is compiled. - PR: https://git.openjdk.org/jdk/pull/9081
Re: RFR: 8288001: SOURCE_DATE_EPOCH should not be set by default
On Mon, 13 Jun 2022 09:47:48 GMT, Severin Gehwolf wrote: >> What do you mean by "build time information"? > > @magicus I think it's `-Xinternalversion` which has different output between > Windows and Linux of the same build. But to me that's a feature not a bug. > From the PR description: > >> With default configuration shows `-Xinternalversion` output same as Windows, >> [...] @jerboaa I agree. If anything, this PR makes me want to remove the option to *not* build reproducible. That way, there will be no discrepancies between platforms. - PR: https://git.openjdk.org/jdk/pull/9081
Re: RFR: 8288001: SOURCE_DATE_EPOCH should not be set by default
On Mon, 13 Jun 2022 07:14:15 GMT, Magnus Ihse Bursie wrote: >> At default configuration, SOURCE_DATE_EPOCH is exported as environment >> variable in SetupReproducibleBuild. Then, gcc is affected of >> SOURCE_DATE_EPOCH environment variable. This value is used only to set >> SOURCE_DATE_ISO_8601 (except below), so I removed "export" for >> SOURCE_DATE_EPOCH in SetupReproducibleBuild. And, at building ct.sym, >> SOURCE_DATE_EPOCH environment variable is needed. So I added setting routine >> if SOURCE_DATE_EPOCH is not set. >> >> This fix works fine. With default configuration shows -Xinternalversion >> output same as Windows, and with --with-source-date configuration shows >> -Xinternalversion output specified timestamp. Would you please review this >> fix? > > What do you mean by "build time information"? @magicus I think it's `-Xinternalversion` which has different output between Windows and Linux of the same build. But to me that's a feature not a bug. From the PR description: > With default configuration shows `-Xinternalversion` output same as Windows, > [...] - PR: https://git.openjdk.org/jdk/pull/9081
Re: RFR: 8288001: SOURCE_DATE_EPOCH should not be set by default
On Wed, 8 Jun 2022 07:47:24 GMT, KIRIYAMA Takuya wrote: > At default configuration, SOURCE_DATE_EPOCH is exported as environment > variable in SetupReproducibleBuild. Then, gcc is affected of > SOURCE_DATE_EPOCH environment variable. This value is used only to set > SOURCE_DATE_ISO_8601 (except below), so I removed "export" for > SOURCE_DATE_EPOCH in SetupReproducibleBuild. And, at building ct.sym, > SOURCE_DATE_EPOCH environment variable is needed. So I added setting routine > if SOURCE_DATE_EPOCH is not set. > > This fix works fine. With default configuration shows -Xinternalversion > output same as Windows, and with --with-source-date configuration shows > -Xinternalversion output specified timestamp. Would you please review this > fix? What do you mean by "build time information"? - PR: https://git.openjdk.org/jdk/pull/9081
Re: RFR: 8288001: SOURCE_DATE_EPOCH should not be set by default
On Wed, 8 Jun 2022 07:47:24 GMT, KIRIYAMA Takuya wrote: > At default configuration, SOURCE_DATE_EPOCH is exported as environment > variable in SetupReproducibleBuild. Then, gcc is affected of > SOURCE_DATE_EPOCH environment variable. This value is used only to set > SOURCE_DATE_ISO_8601 (except below), so I removed "export" for > SOURCE_DATE_EPOCH in SetupReproducibleBuild. And, at building ct.sym, > SOURCE_DATE_EPOCH environment variable is needed. So I added setting routine > if SOURCE_DATE_EPOCH is not set. > > This fix works fine. With default configuration shows -Xinternalversion > output same as Windows, and with --with-source-date configuration shows > -Xinternalversion output specified timestamp. Would you please review this > fix? Thank you for your comments. I understood the goal of reproducible build. But now, ENABLE_REPRODUCIBLE_BUILD is set to false in default configuration. Then I think minimize the effort of SOURCE_DATE_EPOCH when reproducible build is disabled. I wonder why build time information is different from Windows and Linux. - PR: https://git.openjdk.org/jdk/pull/9081
Re: RFR: 8288001: SOURCE_DATE_EPOCH should not be set by default
On Wed, 8 Jun 2022 07:47:24 GMT, KIRIYAMA Takuya wrote: > At default configuration, SOURCE_DATE_EPOCH is exported as environment > variable in SetupReproducibleBuild. Then, gcc is affected of > SOURCE_DATE_EPOCH environment variable. This value is used only to set > SOURCE_DATE_ISO_8601 (except below), so I removed "export" for > SOURCE_DATE_EPOCH in SetupReproducibleBuild. And, at building ct.sym, > SOURCE_DATE_EPOCH environment variable is needed. So I added setting routine > if SOURCE_DATE_EPOCH is not set. > > This fix works fine. With default configuration shows -Xinternalversion > output same as Windows, and with --with-source-date configuration shows > -Xinternalversion output specified timestamp. Would you please review this > fix? No no no. The current design is very much intentional. We must export `SOURCE_DATE_EPOCH` for gcc to be able to see it. I wonder just like Severin: *why* do you think this is a problem? Also, the long term goal is to *only* make reproducible builds. The sole reason it is possible to do `--disable-reproducible-builds` is out of an abundance of caution, if the changes needed for full reproducibility had unintended negative consequences. - PR: https://git.openjdk.java.net/jdk/pull/9081
Re: RFR: 8288001: SOURCE_DATE_EPOCH should not be set by default
On Wed, 8 Jun 2022 07:47:24 GMT, KIRIYAMA Takuya wrote: > At default configuration, SOURCE_DATE_EPOCH is exported as environment > variable in SetupReproducibleBuild. Then, gcc is affected of > SOURCE_DATE_EPOCH environment variable. This value is used only to set > SOURCE_DATE_ISO_8601 (except below), so I removed "export" for > SOURCE_DATE_EPOCH in SetupReproducibleBuild. And, at building ct.sym, > SOURCE_DATE_EPOCH environment variable is needed. So I added setting routine > if SOURCE_DATE_EPOCH is not set. > > This fix works fine. With default configuration shows -Xinternalversion > output same as Windows, and with --with-source-date configuration shows > -Xinternalversion output specified timestamp. Would you please review this > fix? It's not clear **why** you don't want `SOURCE_DATE_EPOCH` set by default. Could you explain? It would be good to make the OpenJDK build reproducible by default. The way it works currently, is a first step towards that goal. - PR: https://git.openjdk.java.net/jdk/pull/9081
