Writing modules

I’m attempting to write a module, the first thing I’m struggling with is a function called module_base.
I’ve found the function in the BatMgrHal sources, but I can’t work out how it is included.
I’ve set my OBJS = asm in the Makefile, but I’m guessing I’m missing something else as it’s ignoring it. I’m still getting (Error) Undefined symbol(s) module_base, referred to from @.o.qfa

One thing that I notice is that the BatMgrHal sources seem to be using a slightly different format of makefile than I’m using, I’ve based mine on the ones out of the DDE example.

Am I better spending some time at the beginning using this different format, is this the “new” one?

Not much detail about what you want to use this for, so I assume you are writing it in C and needing a pointer to the base address of your module?

If so, you will need to generate an assembler file that looks like this:

        IMPORT |Image$$RO$$Base|
        EXPORT module_base

        AREA CodeBase, DATA

module_base
        DCD |Image$$RO$$Base|

        END

In your C program, you do this:

extern int module_base;

It can then be accessed like any other variable (note – variable, not a function).

In your MakeFile, you compile the C code, you assemble the assembler code, and you pass the names of all of the object files (o.blah) in the Link command. !Make (or !AMU) will work out what is necessary, build it if needs be, then paste all the pieces together.

Try also… http://www.chiark.greenend.org.uk/~theom/riscos/docs/CodeStds/ASDTF.txt

Thanks Rick, I’ve already worked out the assembler and the C side of things.
I just needed a kick in the head for the makefile side of things, which now you’ve explained I’ve just done :-D

Obvious I know, but I’m incredibly rusty.

FYI it’s a module to handle (initially) sockets.
I’m trying to keep it all C wherever I can, I did once write a compiler for my own language on my A3000 – but that’s when I was a teenager, and I have regained some (but not most) of my sanity since then.

One side will copy data off the socket buffer into a dynamic memory area. The wimp side of things will then actually deal with the data it’s read.
i.e. you ask it to download a file and specify a location.
The “kernel” side of things will do the dns lookup, open the socket, and read the data into a block of memory.
The “WIMP” side of things will process the actual data, then save the result into a file or hand it off to the process that asked for it.
My hope is that having a lightweight “thread” (it’s not I know) who’s purpose is to empty the network buffer, will make the downloading as fast as it can be.
It doesn’t really matter how long it takes to process the data afterwards or during, as that doesn’t affect download speeds. But if the buffer gets filled up TCP/IP will start throttling the connection.
I’ve noticed using netsurf to download 4 or 5 large files takes an age longer on my pandaboard than on my PC.
I’ve never seen transfers above 1MB/sec using netsurf.
However on my PC I see 5MB/sec and above frequently.
Also FTPc just seems really slow, that could be just me being me, but it doesn’t seem to get near 100Mbit speeds.
But I’m lucky enough to get around 120Mbit download capacity, so I can push the pandaboard to it’s limit in theory – well I’ll use my PC as a server to test it, as then it’s gig speed from PC and repeatable.
I’m hoping to build a kick ass network handling utility.
If I’m wrong and it’s not any better, I’m not bothered as it’s a nice way to learn RISC OS modules. I’ll be disappointed slightly if it’s worse, but either way it will make my life easier for developing other internet applications as I don’t need to deal with sockets etc. again except in the module code, which will hopefully make the applications slightly simpler, as the socket code in NewsUK has not been really easy to code and I wanted to make it far more complex, but it’s too difficult to integrate it and then port it to an existing application, or even to a new one.
Which probably means I’ll get more crazy ideas to make them more complex!

But now I’ve got the thing compiling again, I can continue on my un-relenting quest for pain!

Thanks again, really appreciate the assistance!
And sorry that’s like War and Peace

The “kernel” side of things will do the dns lookup, open the socket, and read the data into a block of memory.

When you start looking you will find the DNS functionality to be somewhat less than it ought to be.
It’s a few years since I looked (back before AntiSpam got a DNS blacklist function) and in my hacking around I noted that reverse look up did some nasty shortcuts meaning it didn’t work properly.
The forward look up isn’t that brilliant either as I recall.

Basically the network stack is elderly and back when it was done it was a quick and dirty effort.
Improve it and you will have a lot friends.

Funny you should mention DNS, it’s the first thing I want to look at. The current lookup method is just not suitable as for starters it’s a sync process, so if DNS is down etc. then it just locks up RISC OS until the timeouts kick in, which as you say is “sub-optimal” ;-)

Yeah I think the forward has bugs with multiple A records.

One thing I was thinking of doing for a laugh was a load-balancer for RISC OS, perhaps call it “Nut-Scraper” ;-)

Funny you should mention DNS, it’s the first thing I want to look at.

Reverse lookup has, IIRC, an assumption of the domain suffix. Deficiencies in the Resolver module were what caused Frank de Bruijn (AntiSpam maintainer) to write an alternate module for that purpose.

As he said a little while back (CSA* or here?) it isn’t a complete resolver, but I suspect it might be a useful starting point.

One thing I was thinking of doing for a laugh was a load-balancer for RISC OS

Use?

Thanks for that I’ll take a look.

Use? “for a laugh” – I thought having a usable load-balancer running on RISC OS maybe a useful edge case say running on a Pi.
Also it will allow me to test latency/throughput a little finer.
But if I’m being honest, largely for the laughs… If I bother it’s not exactly going to be my number one priority ;-)

Priorities are:
1) Socket code
2) HTTP code
3) FTP code
4) DNS code
5) IMAP code
6) POP3 code
7) FTP filesystem
8) NewsUK update
9) WeatherUK update
10) Jabber Client
11) RSS Reader
12) Twitter Client
13) SSL/TLS code
14) Retirement ;-)

Priorities are:
1) Socket code
2) HTTP code
3) FTP code
4) DNS code
5) IMAP code
6) POP3 code
7) FTP filesystem
8) NewsUK update
9) WeatherUK update
10) Jabber Client
11) RSS Reader
12) Twitter Client
13) SSL/TLS code
14) Retirement ;-)

13 probably needs to move up to join item 6 maybe earlier.

Agreed, but I know enough about encryption to know it should probably come after 14, or possibly I should add:
15) Death
16) SSL/TLS code

After a quick google I found this
So it looks fairly straight forward, I’m not really interested in completeness at this stage – just a basic async dns lookup for single A records would be a major step forward for me.

I’ve noticed that there seems to be a preference to using OSLIB rather than toolbox in the modules.
Is this just a preference or is there a problem using toolbox?
I’m hoping it’s a preference thing, as I’ve just got used to toolbox!

Update: Nevermind, getting a module working on the desktop is a waste of my time. I’m still getting Aborts when I try and initialise. I’ll never get the thing written that way.
Far better to have a small, stable module and a separate desktop application I think.
I can always think about a merge later on, or someone else can do it ;-)
Think I’m trying to fly before I can even walk.