[pgrouting-dev] duplicate function names in core

[Stephen Woodbridge]

On 7/10/2012 5:23 AM, Mario Basa wrote:
> Hello,
>
> I am encountering problems placing the drivedist.c and
> boost_drivedist.cpp into core (everything compiles, drivedist works
> but astar searches dumps core. If I take out drivedist, astar works
> like normal.), so I decided to poke around the source. I just found
> out that there are so many duplicate functions (i.e.
> finish(),text2char(),etc.) and struct names (edge_column,etc.) in each
> of the C files.
>
> Since C does not have any namespace functionality nor function
> overloading, I don't have any idea which one will be called whenever
> there is a library call. Scary since this will be difficult to debug.
>
> Now that pgRouting is getting source contributions, I think this will
> be a good time that we start discussing function naming conventions to
> avoid duplicates as well as directory (core and extensions)
> structures.

Mario,

UGH! Ideally one of two things should happen:

1. common functions with compatible signatures should be extracted into 
a common.c common.h or utils.c utils.h and get shared by all algorithms 
that need them.

2. local functions should all be declares "static" which means they are 
not visible or callable from references outside the file they are 
declared in.

So, if you have a C wrapper and it needs functions and no other files 
need to use these functions then they should be declared as static and 
placed in the C wrapper file.

If you have some function that multiple C files need to call or you C++ 
also needs to call, then it needs to go into a file like 
<namespace>_utils.c and all the functions in this file need to have a 
unique <namespace>_functionname() like signature.

I think this will minimize the changes we need to make to the existing 
code using "static" declarations for most things, and then moving the 
others into some namespace'd set of files.

How does this sound?

I like the idea of having multiple library .so files rather than one 
huge file because it keeps things separated but more so because it makes 
it easy to rebuild one library without effecting any of the others. I 
would like to know that if I change TRSP that there is little to zero 
change that I can impact the other algorithms.

Thank about a production environment where you have pgrouting running on 
lots of databases and in one database you want to add TRSP. By making it 
one big library you force a change on all the databases that did not 
need that change, but if TRSP is in a separate library then all the 
database that are not using it can be assured to be stable because they 
did not change. The worsted cases is that the new database with TRSP 
added to it might be unstable and if there is a collision of files like 
you describe the cases crashes, it only happens in the new database. 
Hmmm, maybe it doesn't work this way once the postgresql loads the 
library. But I don't remember having problems loading multiple libraries.

Thoughts?

-Steve