libevent (1.4.12-stable) -> external/bsd

[Iain Hibbert]

Hi,
   I notice that tls@ updated libevent to 1.4.11-stable earlier this year
though within a couple of weeks 1.4.12-stable was released (heh :). I was
looking at updating it and moving to external/bsd at the same time. I've
not got up to installing and testing yet but see the work so far at:

   http://www.netbsd.org/~plunky/libevent-20091028.diff

1. when cross building for sparc64 I find /usr/lib/sparc/libevent* files
are missing from DESTDIR as they don't get installed -- why might that be?

2. I'm not sure how best to deal with the regression tests - they are
currently in regress/lib/libevent and I've switched them to SUBDIR from
regress/lib/Makefile into external/bsd/libevent/regress and reachover from
there but not sure if thats going to work -- suggestions?

3. I 'fixed' the compiler and lint errors by removing the checks that
fail. Its probably not too difficult to fix them properly but will check
with the author and see if they are worth fixing on the 1.4 branch, as 2.x
is in development

4. currently we don't include the "select" subsystem and there is no easy
way to disable it in the configure script so I've added it -- I could work
up a way to discard it if there is a particular reason we don't want it?

iain

[Thor Lancelot Simon-2]

On Wed, Oct 28, 2009 at 02:04:06PM +0000, Iain Hibbert wrote:
> Hi,
>    I notice that tls@ updated libevent to 1.4.11-stable earlier this year
> though within a couple of weeks 1.4.12-stable was released (heh :). I was
> looking at updating it and moving to external/bsd at the same time. I've
> not got up to installing and testing yet but see the work so far at:

Why 1.4.12 instead of 2.x?

I am wondering, though, if we want all the many and bloated components of
libevent -- which is more the case for 2.x than for 1.4.

Thor

[Iain Hibbert]

On Wed, 28 Oct 2009, Thor Lancelot Simon wrote:

> On Wed, Oct 28, 2009 at 02:04:06PM +0000, Iain Hibbert wrote:
> > Hi,
> >    I notice that tls@ updated libevent to 1.4.11-stable earlier this year
> > though within a couple of weeks 1.4.12-stable was released (heh :). I was
> > looking at updating it and moving to external/bsd at the same time. I've
> > not got up to installing and testing yet but see the work so far at:
>
> Why 1.4.12 instead of 2.x?

mostly because Neils' blog mentioned:

  "This is a bugfix-only release, and some of the bugs were kind of nasty.
   I'd recommend that you upgrade, especially if you are writing code that
   uses epoll or evdns"

> I am wondering, though, if we want all the many and bloated components of
> libevent -- which is more the case for 2.x than for 1.4.

I didn't look at 2.x at all but since its still only in alpha, it might be
worth while looking at and possibly providing patches to disable features
we don't want before it turns into a release and we can import it at that
time..

iain

[Thor Lancelot Simon-2]

On Wed, Oct 28, 2009 at 04:28:47PM +0000, Iain Hibbert wrote:

> On Wed, 28 Oct 2009, Thor Lancelot Simon wrote:
>
> > On Wed, Oct 28, 2009 at 02:04:06PM +0000, Iain Hibbert wrote:
> > > Hi,
> > >    I notice that tls@ updated libevent to 1.4.11-stable earlier this year
> > > though within a couple of weeks 1.4.12-stable was released (heh :). I was
> > > looking at updating it and moving to external/bsd at the same time. I've
> > > not got up to installing and testing yet but see the work so far at:
> >
> > Why 1.4.12 instead of 2.x?
>
> mostly because Neils' blog mentioned:
>
>   "This is a bugfix-only release, and some of the bugs were kind of nasty.
>    I'd recommend that you upgrade, especially if you are writing code that
>    uses epoll or evdns"

Didn't I already pick up the only relevant 1.4.12 change by hand?

Thor

[Christos Zoulas-2]

In article <1256738646.364664.28678.nullmailer@...>,
Iain Hibbert  <plunky@...> wrote:

>Hi,
>   I notice that tls@ updated libevent to 1.4.11-stable earlier this year
>though within a couple of weeks 1.4.12-stable was released (heh :). I was
>looking at updating it and moving to external/bsd at the same time. I've
>not got up to installing and testing yet but see the work so far at:
>
>   http://www.netbsd.org/~plunky/libevent-20091028.diff
>
>1. when cross building for sparc64 I find /usr/lib/sparc/libevent* files
>are missing from DESTDIR as they don't get installed -- why might that be?

You need to add stuff in /usr/src/compat/external/bsd/lib and
/usr/src/compat/external/lib

>2. I'm not sure how best to deal with the regression tests - they are
>currently in regress/lib/libevent and I've switched them to SUBDIR from
>regress/lib/Makefile into external/bsd/libevent/regress and reachover from
>there but not sure if thats going to work -- suggestions?

Someone should make them atf?

>3. I 'fixed' the compiler and lint errors by removing the checks that
>fail. Its probably not too difficult to fix them properly but will check
>with the author and see if they are worth fixing on the 1.4 branch, as 2.x
>is in development

ok.

>4. currently we don't include the "select" subsystem and there is no easy
>way to disable it in the configure script so I've added it -- I could work
>up a way to discard it if there is a particular reason we don't want it?

I would not worry too much about it.

christos

[Iain Hibbert]

On Wed, 28 Oct 2009, Thor Lancelot Simon wrote:

> On Wed, Oct 28, 2009 at 04:28:47PM +0000, Iain Hibbert wrote:
> > On Wed, 28 Oct 2009, Thor Lancelot Simon wrote:
> >
> > > On Wed, Oct 28, 2009 at 02:04:06PM +0000, Iain Hibbert wrote:
> > > > Hi,
> > > >    I notice that tls@ updated libevent to 1.4.11-stable earlier this year
> > > > though within a couple of weeks 1.4.12-stable was released (heh :). I was
> > > > looking at updating it and moving to external/bsd at the same time. I've
> > > > not got up to installing and testing yet but see the work so far at:
> > >
> > > Why 1.4.12 instead of 2.x?
> >
> > mostly because Neils' blog mentioned:
> >
> >   "This is a bugfix-only release, and some of the bugs were kind of nasty.
> >    I'd recommend that you upgrade, especially if you are writing code that
> >    uses epoll or evdns"
>
> Didn't I already pick up the only relevant 1.4.12 change by hand?

Perhaps - the only post-import change I saw was the addition of
-fno-strict-aliasing and I didn't examine the rest of the differences
beyond superficially.

Truly there isn't much change but I'm also interested in the
maintainability of pushing it into external/ where a replacement can be
made easily rather than having it spread over the source tree..

iain

[Iain Hibbert]

On Wed, 28 Oct 2009, Christos Zoulas wrote:

> >1. when cross building for sparc64 I find /usr/lib/sparc/libevent* files
> >are missing from DESTDIR as they don't get installed -- why might that be?
>
> You need to add stuff in /usr/src/compat/external/bsd/lib and
> /usr/src/compat/external/lib

Ah right I see it, thanks..

> >2. I'm not sure how best to deal with the regression tests - they are
> >currently in regress/lib/libevent and I've switched them to SUBDIR from
> >regress/lib/Makefile into external/bsd/libevent/regress and reachover from
> >there but not sure if thats going to work -- suggestions?
>
> Someone should make them atf?

Most likely though I don't know how

iain

[Thor Lancelot Simon-2]

On Wed, Oct 28, 2009 at 05:33:07PM +0000, Iain Hibbert wrote:

> On Wed, 28 Oct 2009, Thor Lancelot Simon wrote:
>
> > Didn't I already pick up the only relevant 1.4.12 change by hand?
>
> Perhaps - the only post-import change I saw was the addition of
> -fno-strict-aliasing and I didn't examine the rest of the differences
> beyond superficially.
>
> Truly there isn't much change but I'm also interested in the
> maintainability of pushing it into external/ where a replacement can be
> made easily rather than having it spread over the source tree..

I'm not sure it's a good idea to move it to external if most of the work
will just have to be done again for 2.0.  Most libevent users seem to be
using 2.0 now.

Thor

[Iain Hibbert]

On Wed, 28 Oct 2009, Thor Lancelot Simon wrote:

> On Wed, Oct 28, 2009 at 05:33:07PM +0000, Iain Hibbert wrote:
> > On Wed, 28 Oct 2009, Thor Lancelot Simon wrote:
> >
> > > Didn't I already pick up the only relevant 1.4.12 change by hand?
> >
> > Perhaps - the only post-import change I saw was the addition of
> > -fno-strict-aliasing and I didn't examine the rest of the differences
> > beyond superficially.
> >
> > Truly there isn't much change but I'm also interested in the
> > maintainability of pushing it into external/ where a replacement can be
> > made easily rather than having it spread over the source tree..
>
> I'm not sure it's a good idea to move it to external if most of the work
> will just have to be done again for 2.0.

Well, most of the work is in the move to external/ but I can look at 2.0
afterwards - I don't think we should import an alpha release but it builds
without much trouble and I will look about feeding back the lint and
compiler errors so it will be easier to bring in later.

> Most libevent users seem to be using 2.0 now.

do you have any direct experience of the source compatibility?  The only
references I see in src/*bin are bthset, bthcid, btpand, ftp-proxy and
syslogd and only ftp-proxy is using an 'advanced' API (bufferevent)

iain

[Iain Hibbert]

> > >2. I'm not sure how best to deal with the regression tests - they are
> > >currently in regress/lib/libevent and I've switched them to SUBDIR from
> > >regress/lib/Makefile into external/bsd/libevent/regress and reachover from
> > >there but not sure if thats going to work -- suggestions?
> >
> > Someone should make them atf?
>
> Most likely though I don't know how

Ok, I worked this out and wrote an ATF shell wrapper for the distributed
libevent C test program. Latest diff at

        http://www.netbsd.org/~plunky/libevent-20091030.diff

in summary

        bump libevent to 1.4.12-stable (minor bugfix release)
        move libevent to external/bsd
        include "select" backend
        change regression tests to ATF (all pass on i386)

iain

[Christos Zoulas-2]

In article <1256947698.658358.20461.nullmailer@...>,
Iain Hibbert  <plunky@...> wrote:

>> > >2. I'm not sure how best to deal with the regression tests - they are
>> > >currently in regress/lib/libevent and I've switched them to SUBDIR from
>> > >regress/lib/Makefile into external/bsd/libevent/regress and reachover from
>> > >there but not sure if thats going to work -- suggestions?
>> >
>> > Someone should make them atf?
>>
>> Most likely though I don't know how
>
>Ok, I worked this out and wrote an ATF shell wrapper for the distributed
>libevent C test program. Latest diff at
>
> http://www.netbsd.org/~plunky/libevent-20091030.diff
>
>in summary
>
> bump libevent to 1.4.12-stable (minor bugfix release)
> move libevent to external/bsd
> include "select" backend
> change regression tests to ATF (all pass on i386)

Looks good to me. Are you going to add __RCSID()? Typically we put generated
config.h files in include not in lib, but I don't care either way.

christos

[Iain Hibbert]

On Sat, 31 Oct 2009, Christos Zoulas wrote:

> In article <1256947698.658358.20461.nullmailer@...>,
> Iain Hibbert  <plunky@...> wrote:
> >Ok, I worked this out and wrote an ATF shell wrapper for the distributed
> >libevent C test program. Latest diff at
> >
> > http://www.netbsd.org/~plunky/libevent-20091030.diff
> >
> >in summary
> >
> > bump libevent to 1.4.12-stable (minor bugfix release)
> > move libevent to external/bsd
> > include "select" backend
> > change regression tests to ATF (all pass on i386)
>
> Looks good to me. Are you going to add __RCSID()?

wasn't going to as there were none previously - I modified the
prepare-import.sh script to add $NetBSD$ tags though, is that enough?

> Typically we put generated config.h files in include not in lib, but I
> don't care either way.

Mm, ok - I changed that (also to generate and commit the event-config.h
file after configure as the config.h file won't be required in future)

        http://www.netbsd.org/~plunky/libevent-20091101.diff

(not including the distfiles this time)

iain

[Christos Zoulas]

On Nov 1,  1:58pm, plunky@... (Iain Hibbert) wrote:
-- Subject: Re: libevent (1.4.12-stable) -> external/bsd

| wasn't going to as there were none previously - I modified the
| prepare-import.sh script to add $NetBSD$ tags though, is that enough?

I guess. We can always add them later if we want.

| > Typically we put generated config.h files in include not in lib, but I
| > don't care either way.
|
| Mm, ok - I changed that (also to generate and commit the event-config.h
| file after configure as the config.h file won't be required in future)
|
| http://www.netbsd.org/~plunky/libevent-20091101.diff
|

Fine with me.

christos