The W5100 has the undocumented feature to wrap around the Address Register on an Auto-Increment at the end of physical address space to its beginning.
However, the only way to make use of that feature is to have only a single socket that uses all of the W5100 physical address space. But having only a single socket by defining SINGLE_SOCKET comes with downsides too:
One mustn't call into IP65 network functions anymore after w5100_config(). Additionally the program doesn't support 'W5100 Shared Access' anymore (https://github.com/a2retrosystems/uthernet2/wiki/W5100-Shared-Access).
If the user fails to set the disk tool variable(s) he now doesn't get syntactically wrong commands but rather file-not-founds. Maybe he even chooses to drop tool(s) into the current directory or add the disk tool path(s) to the PATH variable...
So far the base address of the Ethernet chip was a general property of all Ethernet drivers. It served two purposes:
1. Allowing to use a single Ethernet driver for a certain Ethernet chip, no matter what machine was connected to the chip.
2. Allowing use an Ethernet card in all Apple II slots.
However, we now use customized Ethernet drivers for the individual machines so 1.) isn't relevant anymore. In fact one wants to omit the overhead of a runtime-adjustable base address where it isn't needed.
So only the Apple II slots are left. But this should rather be a driver-internal approach then. We should just hand the driver the slot number the user wants to use and have the driver do its thing.
Independently from the aspect if the driver parameter is a base address or a slot number the parameter handling was changed too. For asm programs there was so far a specific init function to be called prior to the main init function if it was desired to chnage the parameter default. This was done to keep the main init function backward compatible. But now that the parameter (now the slot number) is only used on the Apple II anyhow it seems reasonable to drop the specific init function again and just provide the parameter to the main init function. All C64-only user code can stay as-is. Only Apple II user code needs to by adjusted. Please note that this change only affects asm programs, C programs always used a single init function with the Apple II slot number as parameter.
Now that we have per-target combo driver wrappers we can remove the W5100 support from the C64. There isn't any actual (known) W5100-based solution for the C64.
parse_url stores the URL selector in the output_buffer - which is currently 520 bytes. A new entry point called parse_url_buffer was added which instead stores the URL selector in a buffer provided by the user.
url_download now calls the new parse_url_buffer instead of parse_url. The buffer for the URL selector is simply the download_buffer. So the download_buffer is used twice: First to hold the URL selector to be sent as request to the server and then to hold the response received from the server.
However, the URL selector still can't exceed the MSS (aka 1460 bytes).
Note: The User-Agent string was shortened by two bytes as that allows a "default" URL (incl. 'http://' but without port number) of exactly 1400 bytes to end up as 1460 bytes URL selector.
Tweet65 is sample application for triggering an IFTTT maker webhook. If ~450 chars are not enough for the URL then either the IP65 output_buffer needs to be increased or url_parse needs to be modified to use a different buffer. E.g. when called from url_download then the url_download_buffer could be temporarily used to hold the selector.
Beside the code for the actual feature this change introduces analysis of the URL to extract a potential filename to be used for saving. Doing so was a prerequisite for disk download because the filename extension in the URL is the only hint for distinguishing ProDOS sector order disk images from DOS 3.3 sector order images.
So far only the URL input was potentially repeated. Now the file input is potentially repeated as well. For normal file paths there's deliberately still no check. But when entering a drive specifier ("!Sx,Dx") then it is checked for plausibility - and there's an additional confirmation question to keep the user from overwriting the "wrong" disk.
With now having both the URL and file input potentially repeated it became clear that loading and saving the input history shouldn't be repeated - and if the user quits without having entered a valid input the history deliberately isn't saved at all.
So far we were reading (much) more bytes from the W5100 than necessary when processing the HTTP header. The byte were memmove'd to the beginning of the buffer.
However, when downloading a DOS 3.3 sector order disk image (*.dsk / *.do) then we need to place every DOS 3.3 sector at a certain point in the buffer to avoid additional memcpy's later. Therefore the HTTP header processing mustn't read (or rather commit) any body bytes from the W5100. So we now just check for the "\r\n\r\n" after each and every byte. This is of course less than optimal but small/simple - and the header isn't supposed to be that large anyway.
After quite some thoughts it was considered best to simply have the timezone configured in the source code and rebuild if a change is desired/necessary.
IP65 doesn't support TCP flow control. Therefore it doesn't make sense to write a program receiving a significant amount of data via TCP using IP65. From that perspective it makes sense that IP65's HTTP client doesn't allow to handle incoming data with a callback but requires a buffer being able to hold the whole HTTP body.
However, on the Apple II there's the Uthernet II card with its W5100 Ethernet controller chip. That chip has it's own TCP implementation supporting TCP flow control. Therefore the wget65 program uses the W5100 TCP capabilities for the HTTP client.
But even with the W5100 TCP implementation in place IP65 still plays a role for the wget65 program as it desires DHCP and requires (usually) DNS. Both are not supported by the W5100 capabilities.
Nowadays register variables are out-of-style. So if they appear in some source code they can be assumed to be explicitly meant for cc65, so don't ignore them.
- Maybe not all Apple II C programs want to use the special I/O buffer setup.
- All Make-target-specific variables should go at the top of the Makefile as they are user settings.
The Apple II file I/O needs a $400 byte page-aligned buffer for every open file. The default implementation takes those buffers from the heap. However, if a program doesn't use the heap for anything else it is more efficient to place those buffers below the program at $800 by linking iobuf-0800.o and thus avoiding to link any heap code at all. If a program never opens more than one file at a time it can have its start address at $800 + 1 x $400 = $C00.
It may be useful to be able to run the C programs right from the selector (without loading BASIC.SYSTEM first). Or maybe run them on boot by placing the LOADER.SYSTEM as first file in the root directory.
If there's RTC active then the user can at least have the current date set. However, if the machine runs for several days the user needs to re-run Date65.
HttpFileServer65 is inspired by "HFS ~ Http File Server" (http://www.rejetto.com/hfs/). It features an intuitive mapping of local disks and directories to HTTP paths. This allows to not only navigate to some file but to directly enter some file URL. This is especially usefull for downloads via wget, curl or alike.
However, in contrast to HFS it doesn't allow file uploads, only file downloads. On targets with complete DIO support it additionally allows disk image downloads.
Note: The Atari build currently fails as there's so far no getdevicedir() in the Atari C library.
Given that Telnet65's 80 column display requires an enhanced Apple //e anyway it seems reasonable to have the C applications benefit from building for that target too.
The date application isn't of much actual use. But it shows
- how easy to use the IP65 C API is
- the power of combining IP65 with the C library
(going from SNTP time format to a date string in two lines of code)
Lars informed me that $F142 by default produces ASCII code 0 via Ctrl-@. One just needs a way to distinguish it from the no-key-press.
This means that the further processing can be done in that same as on the Apple II - meaning to just replace in the VT100 keyboard translation table $00 with $FF as code for ignore-key.
Chris made me aware that ASCII code 0 actually is a valid character (usually entered as Ctrl-Space or Ctrl-@) and that it is actually used (i.e. by EMACS).
The Apple II allows to natively enter the ASCII code 0 via Ctrl-@. However so far get_key_if_available returned 0 in accumulator to signal that no key was pressed. In order to allow the Apple II get_key_if_available to return the ASCII code 0 in the accumulator I changed it to use the carry flag to signal that no key was pressed.
Because get_key_if_available needs of course to behave the same on all targets I changed the other implementations to use the carry flag too.
Unfortunately I don't know enough about input capabilities of the C64 to decide on how to best get Telnet65 to send ASCII code 0 there.
I would have expected that no matter what causes the TCP connection to be closed would be reported as a TCP call back with lenghth = -1. However there are scenarios where this isn't true. I.e. when the user aborts during a blocking TCP retransmit opration. I'm unsure if this behaviour is to be considered a bug or if this is by some design I don't understand.
So I opted to not change that behaviour but rather add an explicit check for tcp_send returning with an error indicating that the connection is closed. This way the users gets noticed on the closed connection at least on his next key press.
It might be welcome to be able to abort DHCP, DNS and TCP connect operations. However while we're connected to the remote host we need full control over the keyboard. If TCP send operations don't work immediately the user will just have to wait until they do (or the timeout).
On the Apple II the timer_read function is actually a delay function. Therefore we want to avoid calling it if we're busy processing incoming data. Fortunately timer_read is only necessary to trigger a TCP keep alive message. But if we're busy processing incoming data then we for sure need no TCP keep alive. So we simply mark if we just processed incoming data and skip the timeout check.
