On 02/02/17 09:22, Markus Metz wrote:
On Wed, Feb 1, 2017 at 10:46 PM, Vaclav Petras <wenzesl...@gmail.com
<mailto:wenzesl...@gmail.com>> wrote:
On Wed, Feb 1, 2017 at 3:24 PM, Moritz Lennert
<mlenn...@club.worldonline.be <mailto:mlenn...@club.worldonline.be>> wrote:
I don't really agree with the idea that
"Unfortunately, its [GRASS'] development is stagnating because of
small interest
from fresh and young developers. This is partially caused by the fact
that its design and
concepts are overcome by modern practices in a software development."
I do not see GRASS stagnating. And even though GRASS uses an "old"
language,
I think C is fine and that might be more clear now than in 2008, but
C++ is popular too.
C++ is more problematic in terms of portability and robustness over
time, have a look at the revision log of e.g. r.terraflow and the
iostream library.
However, the motivation for GSoC is make writing of modules in C and
C++ easier. We may discuss if "development is stagnating because of
small interest" is true or not, but for sure we can have better
interface which would attract more people.
Then the project should determine what is "better". This could be a
project where a concept is going to be developed without writing a
single line of code. The main objective would then be to first identify
what is bad or too complicated and how to improve on it.
More pressing problem however are the modules which use variants of
all-in-memory and tiling-on-disk raster reading modes. I'm not sure what
is the state now, because MarkusM fixed a lot of those, but some addons
were not included into core because of custom segment libraries (and
even now they have custom all-in-memory implementations).
There are only a few modules requiring random access and external memory
(e.g. but not only tiling on disk). IMHO, random access and usage of
external memory are special cases, and I don't see how a higher-level
API would help with these special cases.
All-in-memory processing is so simple that any API would IMHO make
things more complicated.
The interface to the segment library is pretty simple: you need only
Segment_open(), Segment_put(), Segment_get(), Segment_close().
Maybe it would help if there would a better description about the
differences in the tiling-on-disk methods available in GRASS, i.e. the
segment library, the read-only cache used by r.proj, and the rowio library.
I imagine that it's unpleasant if all you believe in is OO, but that
doesn't necessarily make OO the naturally best way to go... :-)
Similarly to Python API, OOP is not the only thing we should focus on.
C++ is a multiparadigm language and OOP is just part of it (e.g.
templates are kind of big), plus there are different levels of doing OOP
("C with classes" versus more complicated OOP). So yes, the student
should be familiar with more than just OOP.
Note that the vast majority of portability and compatibility issues
arises from Python and C++.
Maybe this proposal addresses two different things that could be kept
separated:
1) some higher-level C and C++ API
IIUC, a higher-level C API would be between the current API and modules.
What functionality is needed at this intermediate level ?
Concerning the C++ API, I'm just a bit afraid that this will lead to
more and more modules in C++ with the possible issues Markus mentions above.
Also, what exactly is meant by:
"Bridging of C++ RAII and try-catch with GRASS GIS C API exit and
optional memory cleaning must be addressed."
This sounds like another attempt to create persistent state in GRASS (I
guess its this discussion [1] coming back ?). Again, I have no problem
with discussing the issue once again, but I don't think this should be
done within a GSoC project without prior discussion and consensus on the
dev-list.
Moritz
[1] https://lists.osgeo.org/pipermail/grass-dev/2014-April/068345.html
_______________________________________________
grass-dev mailing list
grass-dev@lists.osgeo.org
https://lists.osgeo.org/mailman/listinfo/grass-dev
