Hi,

While testing out the memory pool API I encountered a bug in the
documentation for DbMpoolFile::get
(http://www.sleepycat.com/docs/api_cxx/memp_fget.html). In big bold
letters it explains that

"Page numbers begin at 0; that is, the first page in the file is page
number 0, not page number 1."

The C API says the same thing. However, the test program below seems to
argue otherwise. The first page written is 1, and trying to read page 0
fails. I'm running BDB v 4.4.20, x86, Linux.

Did I miss something?
Ryan

#include "db_cxx.h"
#include <cstdio>

static const size_t PAGE_SIZE = 512;

int main() {
DbEnv env(0);
// use a small number of pages
env.set_cachesize(0, PAGE_SIZE*4, 0);
env.set_tmp_dir("tmp");
env.set_msgfile(stdout);
env.open(NULL, DB_CREATE | DB_PRIVATE | DB_THREAD | DB_INIT_MPOOL,
0);


// open a file in the pool, set to delete on finish
DbMpoolFile* pool;
env.memp_fcreate(&pool, 0);
pool->set_flags(DB_MPOOL_UNLINK, true);
pool->open(NULL, DB_CREATE, 0644, PAGE_SIZE);

printf("<<<< Starting stats:\n");
env.memp_stat_print(DB_STAT_ALL);

const char* actions = "wwwwRwRwRRRR";
bool done = false;
db_pgno_t rpn=0, wpn;
void* p;
while(!done) {
switch(*(actions++)) {
case 'r':
case 'R':
pool->get(&rpn, 0, &p);
printf("\n\n<<<< Read page %d:\n", rpn);
// if(p)
pool->put(p, DB_MPOOL_CLEAN | DB_MPOOL_DISCARD);
rpn++;
break;
case 'w':
case 'W':
pool->get(&wpn, DB_MPOOL_NEW, &p);
printf("\n\n<<<< Wrote page %d:\n", wpn);
memset(p, 0, PAGE_SIZE);
memcpy(p, &wpn, sizeof(wpn));
pool->put(p, DB_MPOOL_DIRTY);
break;
default:
done = true;
}

env.memp_stat_print(0);
}

pool->close(0);
env.close(0);
return 0;
}

Another couple points of potential confusion:

1. If you open a pool with a NULL file name it will be automatically
deleted independent of whether the DB_MPOOL_UNLINK flag is set. If
that's a feature it should probably be documented.

2. Calling close() on a temp file's pool handle will throw an ugly
exception. Fortunately unlink() refuses to delete directories! It might
be better to explicitly check for this condition and either silently
ignore it or return/throw a meaningful error message.

Ryan

Update: It turns out that close() only fails if you specify
DB_MPOOL_UNLINK on a temp file.

Ryan

Ryan wrote: