Re: [Bug-apl] shared memory

[Juergen Sauermann]

Hi,

if you look into older versions (say, around SVN 340) of src/Svar_DB.cc of GNU APL then you will
find examples of using shared memory.

Before APs/APserver was introduced, GNU APL used shared memory to store the data of shared
APL variables (⎕SVO and friends). At some point in time that shared memory was replaced by the
process running APserver, which now owns the shared memory (and does not share it anymore).
Variables in APserver memory are now accessed via signals sent over TCP.

If I remember correctly then the reason for changing to APserver was that you couldn't share memory between
different machines. With APserver you can share variables between APL workspaces running on different
computers.

Apart from that, I would say that if shared memory is a dinosaur then it is an extremely fast one. But using it
properly requires a little experience in that field. At least a thorough understanding of semaphores.

/// Jürgen

On 01/27/2017 07:05 PM, address@hidden wrote:

sorry - i found the testcases/AP*     but if there is any thoughts/comments on this any one has - it would be appreciated

I realize shared memory is considered 'dinosaur' 


On Fri, 27 Jan 2017 10:53:21 -0700
address@hidden wrote:

Hi

I have found a lot of discussion threads on shared memory - the switch from /dev/shm/ dir to unix sockets   etc

is there some example code using the new method that someone could pass to me?

I have a synapse library in fpc that can access sockets and wanted to see if conversion of it to apl was beneficial (speed wise, code complexity wise etc)

thanks