Henry B. Hotz wrote:
On Jan 11, 2007, at 3:22 PM, Howard Chu wrote:
For a number of reasons (anyone following the embrace/extend/
extinguish testimony on groklaw?) even when a native Windows API
exists, we've chosen to only support portable APIs if the parallel
exists. Thus, our Windows builds of software use Cyrus libsasl and
OpenSSL, not Windows SSPI. We use OpenLDAP's libldap, not the we-
claim-it's-LDAP-honest API that Windows provides. In general I think
this is the wiser course of action because it lessens our support
burden, among other things.
Doesn't that make it harder to build your products?
Once, maybe. But it's a sunk cost we invested into our build system 7
years ago. In the end it makes support easier.
In the Postgres case wouldn't that mean that you need to first build/
install/configure KfW, and Cyrus SASL before you could build Postgres
for a native Windows client? That seems like a pretty tall order. I
chose SASL instead of GSSAPI for this project because I (mistakenly?)
thought SASL was natively available on all the target platforms.
Granted I don't do Windows myself, but I didn't want to create those
kinds of problems.
For an open source project I would think the support burden of
requiring lots of ancillary installs would be significant.
We create our own build of OpenSSL, BerkeleyDB, libtool, Cyrus SASL,
Heimdal Kerberos, and OpenLDAP on all of our supported platforms. It's
essential to allowing us to provide commercial support for our packages
in a cost-effective manner. All of our software installs in an isolated
filesystem hierarchy, so we have no interference from or dependency on
obsolete versions that may already be present on the host system. It
means we only have to worry about one version of each package, and we
know exactly what build options were used with each package, and we know
that each package was built correctly, and that everything plays
together. (We also offer support to customers using their own builds, or
platform-vendor-supplied builds, but our support rate is double in that
case, because it costs us more time to deal with their unknown environment.)
We have a single master-Makefile that builds everything identically
across Solaris, AIX, HPUX, Linux, Windows, MacOSX, and even z/OS. Our
test suites also run identically on each platform. It would be a pain to
recreate all of it from scratch today, but obviously we've been at this
for a long time.
The burden of requiring lots of ancillary installs is nothing, compared
to the burden of tracking these projects' ongoing development. But since
we depend on all of their functionality, it's what we have to do. I'm an
active contributor on all of these projects, and I know them all inside
and out. That's the real challenge, and our expertise is the real
value-add in our offerings.
--
-- Howard Chu
Chief Architect, Symas Corp. http://www.symas.com
Director, Highland Sun http://highlandsun.com/hyc
OpenLDAP Core Team http://www.openldap.org/project/
