From: Jeremie Courreges-Anglas <jca@wxcvbn.org>
Subject: Re: net/tcpflow: fix build
To: ports@openbsd.org
Date: Sun, 9 Jun 2024 19:04:32 +0200

On Sun, Jun 09, 2024 at 11:03:54AM +0100, Kirill A. Korinsky wrote:
> On Sat, 08 Jun 2024 23:35:32 +0100,
> Kirill A. Korinsky <kirill@korins.ky> wrote:
> > 
> > Probably it worth to add an -extra flavor with all features.
> > 
> 
> Here small update for net/tcpflow which includes:
>  - add textproc/tre to enjoy regexes;
>  - extended flavor to use cairo and exiv2.

tcpflow isn't linked against libexiv2 and its header files aren't
included because HAVE_EXIV2 isn't defined.  portcheck warns about it.
So exiv2 doesn't bring any feature and it doesn't make sense to add
it as a dep.

Regarding cairo support I don't know if it is useful.  What bothers me
is that configure.ac adds a bunch of other libraries because upstream
uses a static link there on mingw.  Self-inflicted pain, and extra
libs look extra.  LIB_DEPENDS=archivers/bzip2 just look bogus, cairo
doesn't depend on it in our tree.

To me this all looks fragile and I wouldn't link a privileged tool to
such libraries.  You don't *have* to ship such a FLAVOR.

The reordering in the Makefile make AUTOCONF_VERSION/AUTOMAKE_VERSION
+ the lone CONFIGURE_STYLE at the end look bizarre.

So that leaves the addition of textproc/tre (I didn't look) and the
need to add --without-o3 because the build currently uses -O3 instead
of respecting CFLAGS.

> I don't enable X11-related stuff by default to allow use it on machines
> without installed x11 sets.
> 
> The diff:
> 
> diff --git net/tcpflow/Makefile net/tcpflow/Makefile
> index 094dcec0696..2e539a179ad 100644
> --- net/tcpflow/Makefile
> +++ net/tcpflow/Makefile
> @@ -3,6 +3,7 @@ COMMENT=		tool for capturing data from TCP connections
>  V=			1.6.1
>  DISTNAME=		tcpflow-${V}
>  CATEGORIES=		net
> +REVISION=		0
>  
>  SITES=			https://github.com/simsong/tcpflow/releases/download/tcpflow-${V}/
>  
> @@ -13,16 +14,26 @@ MAINTAINER=		Kirill A. Korinsky <kirill@korins.ky>
>  # GPLv3
>  PERMIT_PACKAGE=		Yes
>  
> -WANTLIB += ${COMPILER_LIBCXX} c crypto m pcap sqlite3 ssl z
> +WANTLIB += ${COMPILER_LIBCXX} c crypto m tre pcap sqlite3 ssl z
>  
>  COMPILER=		base-clang ports-gcc
>  
>  BUILD_DEPENDS=		devel/boost
> -LIB_DEPENDS=		databases/sqlite3
> -
> -# a test for cairo has a side effect which brokes pcap test
> -# when it can't find cairo, so, disable it.
> +LIB_DEPENDS=		databases/sqlite3 \
> +			textproc/tre
> +
> +FLAVORS=		extended
> +FLAVOR?=
> +
> +# allow using tcpflow on machine without X11 set by default
> +.if ${FLAVOR:Mextended}
> +WANTLIB += bz2 cairo expat fontconfig freetype pixman-1
> +LIB_DEPENDS+=		archivers/bzip2 \
> +			graphics/cairo \
> +			graphics/exiv2
> +.else
>  CONFIGURE_ARGS=		--enable-cairo=false
> +.endif
>  
>  AUTOCONF_VERSION=	2.71
>  AUTOMAKE_VERSION=	1.16
> diff --git net/tcpflow/patches/patch-configure_ac net/tcpflow/patches/patch-configure_ac
> index a9e99810a6d..97353f53303 100644
> --- net/tcpflow/patches/patch-configure_ac
> +++ net/tcpflow/patches/patch-configure_ac
> @@ -1,6 +1,15 @@
>  Index: configure.ac
>  --- configure.ac.orig
>  +++ configure.ac
> +@@ -77,7 +77,7 @@ esac
> + if test x"${mingw}" == "xno" ; then
> +   # Bring additional directories where things might be found into our
> +   # search path. I don't know why autoconf doesn't do this by default
> +-  for spfx in /usr/local /opt/local /sw /usr/local/ssl /usr/boost/include ; do
> ++  for spfx in /usr/local /opt/local /sw /usr/local/ssl /usr/boost/include /usr/X11R6 ; do
> +     AC_MSG_NOTICE([checking ${spfx}/include])
> +     if test -d ${spfx}/include; then
> +         CPPFLAGS="-I${spfx}/include $CPPFLAGS"
>  @@ -490,36 +490,6 @@ AC_CHECK_TYPES([sa_family_t], [], [],
>   ]]
>   )
> 
> 
> -- 
> wbr, Kirill
> 

-- 
jca