Subversion Repositories f9daq

------------------------------------------------------------------------------

vxi11_1.08 - 3/09/2009

Added a sanity check for link->maxRecvSize to make sure it's >0. This gets 

around a bug in some versions of the Agilent Infiniium scope software.

Changed the erroneous strncpy() to memcpy() in vxi11_send, as we could be

sending binary data (not just strings).

Changed a lot of char *'s to const char *'s in an attempt to get rid of

pedantic gcc compiler warnings.

------------------------------------------------------------------------------

vxi11_1.07 - 9/10/2007

Minor change to vxi11_receive_data_block(), this fn now copes with instruments

that return just "#0" (for whatever reason). Suggestion by Jarek Sadowski,

gratefully received.

------------------------------------------------------------------------------

vxi11_1.06 - 31/08/2007

Bug fix in vxi11_receive(), to ensure that no more than "len" bytes are ever 

received (and so avoiding a segmentation fault). This was a bug introduced in 

release 1.04 whilst making some other changes to the vxi11_receive() fn.

Many thanks to Rob Penny for spotting the bug and providing a patch.

------------------------------------------------------------------------------

vxi11_1.05 - 11/07/2007

Added the ability to specify a "device name" when calling vxi11_open_device().

For regular VXI11-based instruments, such as scopes and AFGs, the device name

is usually "hard wired" to be "inst0", and up to now this has been hard wired

into the vxi11_user code. However, devices such as LAN to GPIB gateways need

some way of distinguishing between different devices... they are a single

client (one IP address), with multiple devices.

The vxi11_user fn, vxi11_open_device(), now takes a third argument

(char *device).

This gets passed to the core vxi11_open_device() fn (the one that deals with

separate clients and links), and the core vxi11_open_link() fn; these two

core functions have also had an extra parameter added accordingly. In order

to not break the API, a wrapper function is provided in the form of the

original vxi11_open_device() fn, that just takes 2 arguments

(char *ip, CLINK *clink), this then passes "inst0" as the device argument.

Backwards-compatible wrappers for the core functions have NOT been provided.

These are generally not used from userland anyway. Hopefully this won't

upset anyone!

vxi11_cmd, the simple test utility, has also been updated. You can now,

optionally, pass the device_name as a second argument (after the ip

address). The source has been renamed to vxi11_cmd.cc (from vxi11_cmd.c), as

it is C++ code not C.

Some minor tidying up in vxi11_user.h

With thanks to Oliver Schulz for bringing LAN to GPIB gateways to my

attention, for suggesting changes to the vxi11_user library to allow them to

be accommodated, and for tidying some things up.

------------------------------------------------------------------------------

vxi11_1.04 - 10/07/2007

Patch applied, which was kindly provided by Robert Larice. This sorts out

the confusion (on my part) about the structures returned by the rpcgen

generated *_1() functions... these are statically allocated temporary structs,

apparently. In the words of Robert Larice:

******

Hello Dr. Sharples,

  I'm sending some patches for your nice gem "vxi11_1.03"

  In the source code there were some strange comments, concerning

   a commented free() around ... Manfred S. ...

  and some notes, suggesting you had trouble to get more than one link

    working.

  I think thats caused by some misuse of the rpcgen generated subroutines.

  1) those rpcgen generated *_1 functions returned pointers to

       statically allocated temporary structs.

     those where meant to be instantly copied to the user's space,

       which wasn't done

     thus instead of

       Device_ReadResp  *read_resp;

       read_resp = device_read_1(...)

     one should have written someting like:

       Device_ReadResp  *read_resp;

       read_resp = malloc(...)

       memcpy(read_resp, device_read_1(...), ...)

  2) but a better fix is to use the rpcgen -M Flag

       which allows to pass the memory space as a third argument

     so one can write

       Device_ReadResp  *read_resp;

       read_resp = malloc(...)

       device_read_1(..., read_resp, ...)

     furthermore this is now automatically thread save

  3) the rpcgen function device_read_1

       expects a target buffer to be passed via read_resp

     which was not done.

  4) the return value of vxi11_receive() was computed incorrectly

  5) minor,  Makefile typo's

      CFLAGS versus

      CLFAGS

******

Robert didn't have more than one device to try the patch with, but I've just

tried it and everything seems fine. So I've removed all references to the

VXI11_ENABLE_MULTIPLE_CLIENTS global variable, and removed the call to

vxi11_open_link() from the vxi11_send() fn. There has been an associated

tidying of functions, and removal of some comments.

Thanks once again to Robert Larice for the patch and the explanation!

------------------------------------------------------------------------------

vxi11_1.03 - 29/01/2007

Some bug-fixes (thanks to Manfred S.), and extra awareness of the

possibility that instruments could time out after receiving a query WITHOUT

causing an error condition. In some cases (prior to these changes) this

could have resulted in a segmentation fault.

Specifically:

(1) removed call to ANSI free() fn in vxi11_receive, which according to

    Manfred S. "is not necessary and wrong (crashes)".

(2) added extra check in vxi11_receive() to see if read_resp==NULL.

    read_resp can apparently be NULL if (eg) you send an instrument a

    query, but the instrument is so busy with something else for so long

    that it forgets the original query. So this extra check is for that

    situation, and vxi11_receive returns -VXI11_NULL_READ_RESP to the

    calling function.

(3) vxi11_send_and_receive() is now aware of the possibility of being

    returned -VXI11_NULL_READ_RESP. If so, it re-sends the query, until

    either getting a "regular" read error (read_resp->error!=0) or a

    successful read.

(4) Similar to (2)... added extra check in vxi11_send() to see if

    write_resp==NULL. If so, return -VXI11_NULL_WRITE_RESP. As with (3),

    send_and_receive() is now aware of this possibility.

------------------------------------------------------------------------------

vxi11_1.02 - 25/08/2006

Important changes to the core vxi11_send() function, which should be

invisible to the user.

For those interested, the function now takes note of the value of

link->maxRecvSize, which is the maximum number of bytes that the vxi11

intrument you're talking to can receive in one go. For many instruments

this may be a few kB, which isn't a problem for sending short commands;

however, sending large chunks of data (for example sending waveforms

to instruments) may exceed this maxRecvSize. The core vxi11_send() function

has been re-written to ensure that only a maximum of [maxRecvSize] bytes are

written in one go... the function sits in a loop until all the message/

data is written.

Also tidied up some of the return values (specifically with regard to

vxi11_send() and vxi11_send_data_block() ).

------------------------------------------------------------------------------

vxi11_1.01 - 06/07/2006

Fair few changes since v1.00, all in vxi11_user.c and vxi11_user.h

Found I was having problems talking to multiple links on the same

client, if I created a different client for each one. So introduced

a few global variables to keep track of all the ip addresses of

clients that the library is asked to create, and only creating new

clients if the ip address is different. This puts a limit of how

many unique ip addresses (clients) a single process can connect to.

Set this value at 256 (should hopefully be enough!).

Next I found that talking to different clients on different ip

addresses didn't work. It turns out that create_link_1() creates

a static structure. This this link is associated with a given

client (and hence a given IP address), then the only way I could

think of making things work was to add a call to an

vxi11_open_link() function before each send command (no idea what

this adds to overheads and it's very messy!) - at least I was

able to get this to only happen when we are using more than one

client/ip address.

Also, while I was at it, I re-ordered the functions a little -

starts with core user functions, extra user functions, then core

library functions at the end. Added a few more comments. Tidied

up. Left some debugging info in, but commented out.

------------------------------------------------------------------------------

vxi11_1.00 - 23/06/2006

Initial release.

------------------------------------------------------------------------------