Re: [NEW] Notmuch 0.29.1
On Sat, 23 Nov 2019 at 23:35:08 +0100, Enric Caussa Morales wrote: > Here's an updated version of the port: > > - it's now using the SUBDIR functionality again. The WIP name is "core" > since it's less ugly than having "notmuch/notmuch" in the path. Let me > know if you come up with a more appropriate name. > > - I've incorporated your changes such as the LIBnotmuch... vars and > such. > > - Before reading your mail, i was linking the gnu coreutils into the > port's bin dir so that the tests worked. I guess actually having and > OpenBSD lib like what you suggested is better so that it may be > upstreamed. > > - The library (libnotmuch.so) is now versioned. However, I put the > library version in Makefile.inc so that it might be shared between the > python bindings and the main notmuch package. I was expecting that > including WANTLIB += notmuch in the py-notmuch Makefile would make it > possible to use LIBnotmuch_VERSION, but that is not the case (is there > another way?). > > I've been using it for a day so far and nothing strange happened yet. After > sending this mail, I'll contact the devs about the use-after-free you > commented (it also happens on my machine, but only at the end of the > tests, I guess it's same as yours). > > Thanks a lot for your feedback and your help on porting this. > I tried building this but it bombed out on the configure step: Checking for GMime session key extraction support... ./configure[566]: gpg: not found No. ... *** Error: Could not extract session keys from encrypted message. ... *** Error: The dependencies of notmuch could not be satisfied. You will need to install the following packages before being able to compile ... *** Error 1 in core (/usr/ports/infrastructure/mk/bsd.port.mk:2845 'do-configure') *** Error 1 in core (/usr/ports/infrastructure/mk/bsd.port.mk:2865 '/build/notmuch-0.29.2/.configure_done') *** Error 1 in core (/usr/ports/infrastructure/mk/bsd.port.mk:2550 'all') ===> Exiting mail/notmuch/core with an error *** Error 1 in /usr/ports/mail/notmuch (/usr/ports/infrastructure/mk/bsd.port.subdir.mk:137 'all') It seems it's hard coded to want 'gpg' while security/gpgmime pulls in gpg2. -- PGP: 0x1F81112D62A9ADCE / 3586 3350 BFEA C101 DB1A 4AF0 1F81 112D 62A9 ADCE
Re: [NEW] Notmuch 0.29.1
Here's an updated version of the port: - it's now using the SUBDIR functionality again. The WIP name is "core" since it's less ugly than having "notmuch/notmuch" in the path. Let me know if you come up with a more appropriate name. - I've incorporated your changes such as the LIBnotmuch... vars and such. - Before reading your mail, i was linking the gnu coreutils into the port's bin dir so that the tests worked. I guess actually having and OpenBSD lib like what you suggested is better so that it may be upstreamed. - The library (libnotmuch.so) is now versioned. However, I put the library version in Makefile.inc so that it might be shared between the python bindings and the main notmuch package. I was expecting that including WANTLIB += notmuch in the py-notmuch Makefile would make it possible to use LIBnotmuch_VERSION, but that is not the case (is there another way?). I've been using it for a day so far and nothing strange happened yet. After sending this mail, I'll contact the devs about the use-after-free you commented (it also happens on my machine, but only at the end of the tests, I guess it's same as yours). Thanks a lot for your feedback and your help on porting this. notmuch-191123.tgz Description: Binary data
Re: [NEW] Notmuch 0.29.1
Hi Stuart and ports! I am away from my workstation until late in the month. I will take a look at your feedback when I get back and further work on the port. Cheers
Re: [NEW] Notmuch 0.29.1
On 2019/10/18 00:08, Enric Morales wrote: > Forgot to attach the updated port port. Here it is. Here's an updated version with cleaner Makefile/patches, there was a COMMENT-python3 in your Makefile but no other parts relating to that so I've left out the python bindings for now and we can add them as a separate port after we're done with this one. Some of the test failures are a bit concerning though, I see a bunch of \xdf printed, this typically indicates a use-after-free (OpenBSD's malloc overwrites some freed chunks with 0xdf characters).. notmuch.tgz Description: application/tar-gz
Re: [NEW] Notmuch 0.29.1
Forgot to attach the updated port port. Here it is. notmuch,5.tgz Description: GNU Zip compressed data
Re: [NEW] Notmuch 0.29.1
On 2019/10/17 22:56, Enric Morales wrote: > Stuart Henderson writes: > > > On 2019/10/17 03:57, Anthony J. Bentley wrote: > >> I'm inclined to move the Python bits to a separate port, so you can > >> use the Python module without running setup.py manually. > > > > That's definitely my preference. > > > > Enric did actually already do this but it seems to be lost from the current > > tar... > > > > https://marc.info/?l=openbsd-ports=155449699431943=2 > > Hi Stuart, > > I think I'll end up doing two separate ports, but if you think that > approach was a good idea, I'll do the subdir stuff instead. > > Cheers, > > Enric Using the subdir like this to share things is still two ports - just means that common parts can be shared so it's harder to forget to keep them in sync. I have a feeling we may have been tied up with 6.5 release when you sent that out before hence it getting forgotten but that was looking pretty close to ready.
Re: [NEW] Notmuch 0.29.1
Stuart Henderson writes: > On 2019/10/17 03:57, Anthony J. Bentley wrote: >> I'm inclined to move the Python bits to a separate port, so you can >> use the Python module without running setup.py manually. > > That's definitely my preference. > > Enric did actually already do this but it seems to be lost from the current > tar... > > https://marc.info/?l=openbsd-ports=155449699431943=2 Hi Stuart, I think I'll end up doing two separate ports, but if you think that approach was a good idea, I'll do the subdir stuff instead. Cheers, Enric
Re: [NEW] Notmuch 0.29.1
Hi Stuart, Anthony and ports@ I guess I didn't run into any problem because I was building it in the same machine that i was running the program on. I created a VM to build packages and, indeed, I noticed some dependencies missing. Besides adding more build depends: - the libnotmuch.so library is now versioned. Sorry, I missed that last time :P - Test depends are taken care of - Tests are now running by default - HTTPS in the HOMEPAGE - Port still using lang/python since AFAIK it needs it for the documentation - The tests must be run with debugging symbols enabled, so I added that to the MAKE_FLAGS. They also need certain GNU coreutils functionality. Things that are weird: - portcheck suggests that the mail/gmime30 is not needed - generating the emacs binding fails unless emacs with gtk2 flavour is installed - The test #750 is the one that hangs halfway. If you switch to the tests directory and run the command that makes it hang you'll see why: NOTMUCH_CONFIG=/usr/ports/pobj/notmuch-0.29.1/notmuch-0.29.1/test/tmp.T750-gzip/notmuch-config ktrace -d ../notmuch show --format=raw id:msg-007@notmuch-test-suite Thank you so much for your help and comments, I love reading your feedback and learning so much. Enric
Re: [NEW] Notmuch 0.29.1
On 2019/10/17 03:57, Anthony J. Bentley wrote: > I'm inclined to move the Python bits to a separate port, so you can > use the Python module without running setup.py manually. That's definitely my preference. Enric did actually already do this but it seems to be lost from the current tar... https://marc.info/?l=openbsd-ports=155449699431943=2
Re: [NEW] Notmuch 0.29.1
On 2019/10/17 00:51, Enric Morales wrote: > I had to add the attached file to the tests/ dir so that the tests > more or less worked. Don't do all the pkg_info -q tests, those should be handled by TEST_DEPENDS > > I still use notmuch daily (using the Emacs frontend right now) and it > works fine for my usecase. > > Comments?
Re: [NEW] Notmuch 0.29.1
Hi Enric,
Thanks for sending this out.
- There's a missing dependency on mail/gmime30.
- Homepage can be switched to https.
- SHARED_LIBS numbering starts at 0.0.
When I try building, it fails right away:
===> Building for notmuch-0.29.1
cat: ./version: No such file or directory
cc -c -DNOTMUCH_VERSION= -std=gnu99 -O2 -pipe -gcommand-line-arguments.c
-o command-line-arguments.o -MD -MP -MF .deps/command-line-arguments.d
command-line-arguments.c:4:10: fatal error: 'error_util.h' file not found
#include "error_util.h"
^~
1 error generated.
gmake: *** [Makefile.local:201: command-line-arguments.o] Error 1
> lang/python/${MODPY_DEFAULT_VERSION_2}
Not the way we do it. lang/python goes in MODULES and we set
MODPY_VERSION (but MODPY_DEFAULT_VERSION_2 is the default, so that's
not needed).
I'm inclined to move the Python bits to a separate port, so you can
use the Python module without running setup.py manually.
Enric Morales writes:
> A couple of the tests that involve the zlib fail:
> Notmuch calls a zlib function and then i start seeing a repeated call to
> getentropy that heats up the system until I kill the program.
Hm, that's something I'd really like to hear more about.
--
Anthony J. Bentley
Re: [NEW] Notmuch 0.29.1
I checked what could be wrong with the tests failing, managing after some tweaks to get 90%+ of the test passing thanks to the GNU Coreutils and the GDB in ports. A couple of the tests that involve the zlib fail: Notmuch calls a zlib function and then i start seeing a repeated call to getentropy that heats up the system until I kill the program. I had to add the attached file to the tests/ dir so that the tests more or less worked. I still use notmuch daily (using the Emacs frontend right now) and it works fine for my usecase. Comments? test-lib-OPENBSD.sh Description: test-lib-OPENBSD.sh
Re: [NEW] Notmuch 0.29.1
Hi ports@, the attached tarball updates the work-in-progress port to upstream notmuch 0.29.1. The notmuch project fixed some build-related issues, fixed bugs and improved the Emacs bindings, now distribute signed sha sums among other changes. See [1] for a detailed list of changes. I've dropped the mentions on Python 2.7 due to Python 2 being sunsetting. The port now builds the bindings for Python 3 and Emacs. I've been using the port for the last week on -CURRENT#366 (2019/10/9 snapshot) and it works fine (I am composing this mail using Emacs+Notmuch). Admittedly, the tests are broken, I might look into them though, as they could provide pointers on what to work on. Please let me know what I can do to improve this port. Cheers, Enric notmuch,4.tgz Description: Notmuch port milestone 4 [1]: https://notmuchmail.org/pipermail/notmuch/2019/028080.html
