After interim review meeting of Exploratoy Software Project, we (ukai, eto, mikio) went to have a dinner at Korean Home cooking, Kayagum at Sugamo, Tokyo. We ate too much...

Today, I have a interim review meeting of my projects in Exploratory Software Project 2004 2nd half, at Bunkyo Green Court 18F.

Mikio Hirabayashi's project, Hyper Estraier: a full-text search system for communities. Core API and Thread API already implemented. Currently designing Node protocol and Node API so start implementing Node server, Node API and Node Farm. Progress 50%. About 1 hour for presentation and Q&A.

Kouichirou Eto's project, qwikWeb - integration of mail, web and meeting. Time series view of page modification was implemented. This idea will be extended as past and future event view so that we can use it as historical view and/or as planning view (scheduler). It would be nice if it can be used to coordinate meeting schedule and records easily. About 1 hour for presentation and Q&A.

Unfortunately, Kazuhisa Takei can't attend today's review meeting because his father is in critically ill.

After that, we discuss more.

someone implements postgresql module using hyper estraier APIs pgestraier , so that we can query to postgresql by using SQL for text indexed by hyper estraier.

Apache module mod_estraier that registers index of contents handled by Apache and provides search web page would be killer application of hyper estraier. It would be easily integrated with Apache and if configured with apache proxy, we can index all contents through the proxy, that is, user can search any browsed contents. apache2-mpm-prefork would be problem, because hyper estraier could be used in single writer and multi reader.

qwik.jp currently broken, it says unregistered FQDN in HELO, so some SMTP, at least debian.or.jp, refuses the connection. It is fixed.

I've noticed that qdbm 1.8.27-1 has been failed to be built on hppa.

hppa-linux-gcc -Wall -ansi -pedantic -fPIC -fsigned-char -O2 -fomit-frame-pointer -DNDEBUG -o dpmgr dpmgr.o -L. -L/home/buildd/lib -L/usr/local/lib -lqdbm -lz -lpthread -lc
/usr/bin/ld: dpmgr: hidden symbol `$$dyncall' in /usr/lib/gcc-lib/hppa-linux/3.3.6/libgcc.a(_dyncall.oS) is referenced by DSO
collect2: ld returned 1 exit status
make[1]: *** [dpmgr] Error 1

I suspected this was because libqdbm.so is linked by /usr/bin/ld but missing required libraries (libgcc or so). I saw the sympotom on warning messages at linking time such as "warning: undefined reference to `$$remI'".

So, I modified Makefile like this:

@@ -421,7 +422,7 @@
 libqdbm.so.10.0.0 : $(MYLIBOBJS)
        if $(LD) -V | grep "GNU" > /dev/null ; \
          then \
-           $(LD) -shared -soname libqdbm.so.10 -o $@ $(MYLIBOBJS) $(LIBLDFLAGS) ; \
+           $(CC) -shared -Wl,-soname,libqdbm.so.10 -o $@ $(MYLIBOBJS) $(LIBLDFLAGS) ; \
          else \
            $(LD) -G -h libqdbm.so.10 -o $@ $(MYLIBOBJS) $(LIBLDFLAGS) ; \
          fi

Then, I could successfully build qdbm 1.8.27 on hppa. Before I uploaded it as 1.8.27-2, I just tried to run "make check" to see whether the binaries work fine or not. Unfortunately, I've got the error. "crtest" claimed the created database was broken.

I consulted with upstream author Mikio Hirabayasi. He confirmed this on HP TestDrive. He wondered this would be the similar sitution with OpenBSD, that won't work with mmap(2) usage of qdbm, so that it requires NOMMAP configuration. qdbm mmaps hash table which exists at the head of the datafile and reads/writes data entries on the rest of the datafile. On OpenBSD, qdbm writes data entry but sometimes zero-filled page was written on the datafile. So, he suggests to disable mmap on linux/hppa.

I also asked gotom. He tried this on another hppa machine (at ishikawa's home) but he got no problem. His hppa kernel version is 2.4.20-pa35. My hppa kernel version is 2.4.19-64, and HP TestDrive hppa kernel version is 2.4.17-64.

Anyway, these are somehow old version of kernel. I upgraded to 2.4.27-64

# apt-get install kernel-image-2.4.27-64
# palo
# shutdown -r now

and the problem has been disappered! Hmm, this problem should be the bug in old kernel, not the bug of qdbm.

I uploaded 1.8.27-2 that fixes FTBFS on hppa.