Previously, there were bare "return;" statements in some places, even though main() was declared as returning int. This is disallowed by C99 and later, and ORCA/C now complains about it when using full lint checks.
It is hard-coded to work with session numbers 1-8, and it may trash memory when passed higher session numbers. This was causing crashes after disconnecting in some cases.
This patch just bypasses the attention vector routine completely for DSI sessions (it's still called for sessions 1-8, i.e. ASP sessions). This should be OK, because its only real function was to display alerts about the connection being closed.
*The OS X server doesn't support ProDOS-type file type info, so trying to set it will give an error. This normally causes GS/OS to give an error when creating a file. There is now an option to ignore the error and proceed with creating the file, although the file type still won't be set.
*The OS X server seems to restrict each write operation to a small size based on the ASP quantum, so there is now an option for this. Using this option means we may write less than requested, but the AppleShare FST can detect this and proceed to write the rest.
On the OS X 10.5 server, FPZzzzz messages seem to work even with a connection negotiated as AFP 2.0, so it's useful to allow this combination. (On Netatalk, the sleep requests are ignored and give an error under AFP 2.0, but they don't kill the connection either.)
This tells the server to keep the session alive (typically for at least several hours) without sending or expecting tickles. This is useful in cases where our run queue procedure is not run regularly or at all, e.g. in text-mode shells.
The implementation here is geared toward compatibility with Netatalk and differs from Apple's published standards in a couple ways. It may or may not work with other servers.
It does this when the ASP layer gives a network error. To work around it, we count the number of sessions (possibly overestimating) and convert the error to "unable to open session" if PFI has not reached its session limit (8).
We indicate that AFPBridge is installed by creating a message in the message center (which also includes its version, in case we want to distinguish between different versions in the future).
This is necessary because FIListSessions2 has a bug that causes it to return garbage in the server/zone fields for DSI sessions. (The bug is essentially that PFI is indexing into its own tables based on the ASP/DSI session number, rather than properly using a PFI-level session index.)
This avoids any possible strange behavior with calling Marinetti in P8 mode. It also ensures that I/O operations in P8 mode will give an error, but not terminate the connection: it will still be alive when we get back to GS/OS, unless the server has killed it due to inactivity (which is quite possible).
This can occur when control-reset is pressed in P8 mode. It’s supposed to reinitialize the AppleTalk stack. In our case, we mark connections to be reset once we're back in GS/OS, since we can't call Marinetti in P8 mode.
This only works if our init is loaded before Marinetti’s, because otherwise Marinetti’s shutdown notification procedure gets control first and shuts down the connection.
In that case, LoadOneTool returns an error, but the tool stub is still loaded, which is good enough (we just need its FPT to be loaded and installed in the TPT).
This should add support for a greater variety of servers, which may only support AFP 2.2 and up (or refuse to use older versions over TCP).
The IIgs software is designed for AFP 2.0. AFP 2.2 is close enough that things seem to work, but there could potentially be a few compatibility issues.
This could be unsafe for some forms of client code, but it looks like the AppleShare FST always actually has a large enough buffer to hold the full amount being requested, and can deal with up to 65535 bytes being returned at once. We take advantage of this to do larger reads in a single request, which provides a noticeable performance advantage.
This code tries to call EasyMount using the request procedure documented in its ERS, but it turns out that doesn't work. The request procedure trashes its return address on the stack, so it crashes whenever you call it.
Stephen Heumann
Stephen Heumann