I think this is the log call that has been killing backends by sending
very large strings.

It's not a good idea to set this option.

Log message less choppy.

There was an info-level log call when a transaction was blocked, but
not a complementary one when the transaction was resumed.  Added the
latter, so that the log provides a clear record of whether the server
is wedged or working.

I was always uncomfortable with this, but its portability is a simple
justification for removing it.

        standard output unless the "EVENT_LOG_FILE" or "STUPID_LOG_FILE"
        environment variable is set.

      - The much-hated name "STUPID_LOG_FILE" now has a preferred
        alias:  "EVENT_LOG_FILE".

turn on TCP_NODELAY, in both directions. This avoids latency when a request or response doesnt fit in one packet. ZEO is already doing its best to buffer data before calling send, so TCP_NODELAY can do little harm

may not believe the bucket is an object" pain.  Simplified the exit maze.

rearranged so that damaged info isn't left behind in case of a
persistence error; added another assert for "the other way" an empty
BTree may get created.

fine; just keeping the test suite as closely in synch as possible.

is empty.

use it, but this is really a step on the way to speeding "lopsided
merges".

macros from Zope3.  These allow converting some runtime tests into
compile-time specialization, and that's going to be more important for
work on speeding "lopsided merges".

weightedIntersection.

intersection and union functions:  get the concept of weight entirely out
of the internal set iteration protocol.  The caller wants values, or
doesn't, and that's just a Boolean.  This makes for some nice little
internal simplifications, but doesn't yet change any semantics.

leaving negative weights alone for now.

are sets.  This just explains what the code actually does; it doesn't
make sense to me, so if it does to you feel encouraged to elaborate (in
the ZCTextIndex code, we never found a use for the weight outputs of
the weighted set ops).

not sure it's telling the full truth yet.

and started repairing obvious errors in their docs.  More to come.

supplied by the correct BTree modules.

the algorithm in a way I believe Jim suggested a few weeks ago (but that
I didn't understand then).

reason for it not being enabled before.

Workaround for bug in Python versions upto and including 2.1.3.  If
the interpreter was builtin without large file support but the
platform had it, os.path.exists() would return false for large files.
The fix is to try to open the file first, and decide whether to create
a new file based on errno.

(Only the two of the last three chunks of the diff affect this
behavior.)

Also cleanup imports and add some whitespace between arguments to func
calls.

one bucket.

returning X, contradicting the docs and common sense.
difference(None, X) continues to return None.

Noted that the way difference() treats None doesn't match the docs and is
almost certainly wrong.

nextBTreeItems() and nextTreeSetItems():  ditto.

the recursion (iteration is quite natural here -- it was tail-recursive).

self, which effectively turned the error into a fatal error.  Fixed.

code, are bogus when PERSISTENCE is #define'd.  Repair them.

a BTree is empty if and only if its firstbucket pointer is
NULL, so there's no need to keep checking nonzero inside the
loop.  Also, the pointers here are all local, so there's no
need to incref and decref them -- the apparent necessity was
just an artifact of using the bucket assignment macros.  So
stopped using those macros, and the need for incref/decref
operations went away too.

PyVar_AssignB(), ASSIGNB(), ASSIGNBC():  Turned out these
are no longer used anywhere anymore, so removed them.

PER_UNUSE():  New macro to capture the endlessly repeated
PER_ALLOW_DEACTIVATION + PER_ACCESSED pair.  As the comment
says, so sue me <wink>.

Bucket_deleteNextBucket():  Documented & simplified.