Test serialization for serial and shared triangulations #15649
Conversation
|
/rebuild |
pcafrica
force-pushed
the
serialization_test
branch
2 times, most recently
from
995e99e
to
dbea6d7
Compare
|
@pcafrica After you rebase, I'll take a look at this PR. |
pcafrica
force-pushed
the
serialization_test
branch
from
dbea6d7
to
e66fbf7
Compare
|
Thank you @peterrum! I have just rebased, this MR is ready for review. |
| #include "./tests.h" | ||
| #include "../grid/tests.h" |
There was a problem hiding this comment.
The tests.h file has just been moved from fullydistributed_grids to grids, without modify its content.
Is it fine to reference it also from other folders than grids?
| { | ||
| boost::archive::text_oarchive oa(stream); | ||
| oa << triangulation; | ||
| oa << vector; |
There was a problem hiding this comment.
I don't think we need to test here the vector. There should be other tests whether the serialization/deserialization of vectors work.
| { | ||
| const std::string filename = "save_load_" + std::to_string(dim) + "d_out"; | ||
|
|
||
| triangulation.save(filename); |
There was a problem hiding this comment.
But here we should test vectors sometime in the future ;) What is missing for that?
There was a problem hiding this comment.
Nothing is missing, indeed :)
This is actually tested in grids/serialization_01.cc.
| std::stringstream stream; | ||
| { | ||
| boost::archive::text_oarchive oa(stream); | ||
| oa << triangulation; |
There was a problem hiding this comment.
I am wondering what happens for p:s:T for load/save? I guess it should work but users need to provide unique file names on each process!?
There was a problem hiding this comment.
Given the current implementation of T::{save,load}(), the following two are equivalent:
boost::archive::text_oarchive oa(std::ofstream(filename));
oa << triangulation;
and
triangulation.save(filename);
I decided to stick with the first one, as it is more versatile (we can easily serialize more information, such as vectors, onto the same file). Also, @blaisb and I are planning to use it for step-83.
The p::s::T::{save,load}() methods were already tested (see sharedtria/tria_load_01.cc), and #15576 did not break it.
So if run in parallel then yes, unique file names must be provided on each process.
There was a problem hiding this comment.
Regarding step-83: I think triangulation.save(filename); is the right approach.
|
@pcafrica One thing I am wondering of is whether |
pcafrica
force-pushed
the
serialization_test
branch
from
e66fbf7
to
d68d963
Compare
Personally, I haven't tested it. From what I see here, there are a couple of I'll try out, and possibly open a new PR. |
|
Fixed, thank you! |
pcafrica
force-pushed
the
serialization_test
branch
from
8b1b0ab
to
65a5df8
Compare
pcafrica
force-pushed
the
serialization_test
branch
from
65a5df8
to
7e9d3d4
Compare
|
@peterrum Do you approve now? |
| std::stringstream stream; | ||
| { | ||
| boost::archive::text_oarchive oa(stream); | ||
| oa << triangulation; |
There was a problem hiding this comment.
Regarding step-83: I think triangulation.save(filename); is the right approach.
Depends on #15576.
Tests for serializing/deserializing were already implemented for
p::d::Tandp::f::T.The new design introduced in #15576 unlocks the same interface also for
Tandp::s::T.This PR adds 3 new tests for serializing/deserializing:
@peterrum @blaisb @bangerth