| Age | Commit message (Collapse) | Author |
|---|
|
This has quite a few changes due to the low-level nature of this code:
when the contents of a string get copied by C functions, we can't pass
it to arguments of the scheme-pointer type directly anymore.
Instead, strings are now wrapper objects which point to an internal
(NUL-terminated!) bytevector. So not only do we have to extract this
bytevector and pass it to these C functions, but we also need to
calculate the string length excluding the NUL terminator. This
creates a bit more branching in the code to get the thing to copy and
length.
The nice thing is that we no longer need to append NUL bytes to
strings in a few places. Instead, we can just pass the raw bytevector
to the C functions, which means this should be rather faster now.
|
|
The WITH(OIDS=true) attribute for CREATE TABLE is no longer supported
since version 12. So simply don't run the test for inserted-oid.
I *think* this means the inserted-oid procedure is completely useless
now, but I'm not 100% sure.
The pg_type table has grown a "oid" column in 9.3 which comes before
the "typename" column, which used to be the first. So the test is a
bit too tightly coupled to the postgres system catalog. Move the test
to where we create actual tables and use it on that, instead.
|
|
|
|
Reported by Kon Lovett
|
|
|
|
This also removes the need to add an Emacs modeline.
Pointed out by Ricardo G. Herdt, as the build didn't work on Alpine.
|
|
|
|
We used to use srfi-18, but that's an extra dependency we don't need
so when we added our own thread waiter with support for a delay, we
dropped srfi-18 altogether, also for calls not using the delay.
Tests did not catch this, ironically, because they load srfi-18.
|
|
This breaks in newer CHICKEN versions.
|
|
|
|
|
|
|
|
This really is just a matter of reading out notifications when they're
available. Also, we offer a way to explicitly wait for notifications.
|
|
|
|
|
|
|
