On 2020/08/27 12:11, Rafael Sadowski wrote:
> Runs very stable for me and speeds up my work immensely.
> 
> Opinions? OKs?

> Index: Makefile
> ===================================================================
> RCS file: /cvs/ports/mail/neomutt/Makefile,v
> retrieving revision 1.56
> diff -u -p -u -p -r1.56 Makefile
> --- Makefile  13 Aug 2020 18:22:58 -0000      1.56
> +++ Makefile  24 Aug 2020 11:00:09 -0000
> @@ -5,7 +5,7 @@ COMMENT=      tty-based e-mail client, Mutt w
>  GH_ACCOUNT=  neomutt
>  GH_PROJECT=  neomutt
>  GH_TAGNAME=  20200626
> -REVISION=    1
> +REVISION=    2
>  
>  CATEGORIES=  mail
>  HOMEPAGE=    https://neomutt.org/
> @@ -16,9 +16,12 @@ MAINTAINER=        Stuart Henderson <stu.ports@
>  PERMIT_PACKAGE=      Yes
>  
>  # uses pledge()
> -WANTLIB+=    c crypto curses iconv idn2 intl ssl tdb z
> +WANTLIB +=   c crypto curses iconv idn2 intl kyotocabinet ssl tdb
> +WANTLIB +=   z
> +
>  
>  LIB_DEPENDS+=        databases/tdb \
> +             databases/kyotocabinet \
>               devel/gettext,-runtime \
>               devel/libidn2
>  BUILD_DEPENDS+=      devel/gettext,-tools \
> @@ -41,6 +44,7 @@ CONFIGURE_ARGS+= --mandir="${PREFIX}/man
>               --fmemopen \
>               --ssl \
>               --tdb \
> +             --kyotocabinet \
>               --debug
>  
>  CONFIGURE_ENV=       autosetup_tclsh="${LOCALBASE}/bin/jimsh" \
> 

I switched from qdbm (for some arches) and kyotocabinet (for others) to
tdb with the update to 20200501, partly to unify things between arches
(kc isn't available on all arches), partly because some of the test
failures in kc make me not really trust it on an OS like OpenBSD that
doesn't have coherent mmap/file access ("unified buffer cache"), and
partly because in benchmarks (hacked up contrib/hcache-bench script,
which times importing a maildir with various different hcache backends
and then reopening it) the tdb backend is typically faster.

What sort of speed difference are you seeing between the backends?

If it's a big improvement then we could readd --kyotocabinet on those
arches which have it but if not then I'd rather avoid the extra dep.

(neomutt also supports lmdb hcache backend, but in my tests that's quite
a lot slower than tdb or kyotocabinet).

Reply via email to