VTK Sequential write

[ligazetom]

Description VTK file writes are all parallel at the moment. Which is nice, but I have some concerns about file server not being able to handle requests from multiple processes. So I wanted to implement/force sequential write of these files. Problem is the only exposed function for writing cell data in p4est_vtk.h is "p4est_vtk_write_cell_dataf" which takes variadic arguments. Since I would like to wrap this sequential functionality to another function, it has to support variadic arguments as well, but then I can pass them forward only as va_list.

In p4est_vtk.h there is another function "p4est_vtk_write_cell_data" which would work, however it is not implemented. In p4est_vtk.c is function "p4est_vtk_write_cell_datav" which is exactly what I need, but it is not exposed.

Obviously I can do whatever I need since I have the source code, the thing is if I provide someone else with only my project source, they have to make the change to the library as well, which is not ideal.

EDIT: The quickest fix would be to make copy of "p4est_vtk_write_cell_dataf" function that takes va_list already.

[cburstedde]

Thanks, those are all good points. If you'd like to make a PR against the develop branch, that would be great.

Exposing p4est_vtk_write_cell_datav sounds good as well as implementing p4est_vtk_write_cell_data. When in doubt about the prototypes, just let me know. Please be generous with doxygen comments.

Doing the same for point data would be really grand.

[ligazetom]

I went with copy of p4est_vtk_write_cell_dataf where I changed variadic arguments for va_list. I thought about exposing the p4est_vtk_write_cell_datav, however there are some checks in p4est_vtk_write_cell_dataf which would become unnecessary responsibilities for the user.

However I cannot think of a good name for this "overload", so if you agree with this solution and can think of something better than p4est_vtk_write_cell_dataf_va_list I'd gladly create PR.

---

You mentioned "doubt about prototypes". I didn't want to create new issue for something like this, but it's been bothering me for a while. I don't understand why none of refine callbacks (regular or _ext) are getting void *user_data.

Obviously it can be worked around using global variables or p4est_t's custom data but that is far from clean. I thought since callbacks for p4est_iterate have such functionality there must be some reason others don't.

[cburstedde]

Thanks for looking into it!

I went with copy of p4est_vtk_write_cell_dataf where I changed variadic arguments for va_list. I thought about exposing the p4est_vtk_write_cell_datav, however there are some checks in p4est_vtk_write_cell_dataf which would become unnecessary responsibilities for the user.

I'd prefer not duplicating any code. Just exposing _cell_datav and adding a bit more documentation on the va_list contents seems fine to me.

However I cannot think of a good name for this "overload", so if you agree with this solution and can think of something better than p4est_vtk_write_cell_dataf_va_list I'd gladly create PR.

Exposing the _cell_datav function, we won't need another name. :)

You mentioned "doubt about prototypes". I didn't want to create new issue for something like this, but it's been bothering me for a while. I don't understand why none of refine callbacks (regular or _ext) are getting void *user_data.

It's historic. User context can be put into the p4est->user_pointer field, which works quite well.

[cburstedde]

Looks great then!

We're still missing the point analogons for data and datav. Contributions welcome.

[cburstedde]

Closing this for now.