Branch v1.6.1 test PR #1
Closed
Conversation
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
It's an optimization when it's on, but whether it *should* be on can vary from processor to processor, and it's a bug (although admittedly not a numerically huge one) when it's off.
When the underlying type here isn't one of our or the standard library's vector types, we don't have an index_range() overload for it. Thanks to @jwpeterson for the fix
`Eigen::SparseMatrix::setZero` clears the entire non-zero pattern, so we have to re-reserve after zeroing in order to not have very expensive recalls to `Eigen::SparseMatrix::reserveInnerVectors`
Assigning to the (template typed) values in these data structures with
the (integer) zero doesn't work when the template type is not POD nor
has an overloaded assignment operator taking an int/number. Assigning
"{}" instead works for a larger set of types, while still giving zero
values to POD. This issue was run into when I tried to use an
eigen::Matrix<DualNumber<...>> in TypeVector.
These functions are called from the ReferenceCountedObject destructor and move constructor, respectively, two functions which generally are not allowed to throw. We should therefore satisfy this requirement by catching any exceptions generated in these functions and immediately calling std::terminate() with a simple error message instead.
Not sure why this was even compiling before, but I do know that it wasn't giving us useful information when it used to evaluate as "true", and it's killing me on a new system where it evaluates as "false".
We had a potential deadlock before as reported by helgrind. We would acquire the `remote_elem_mtx` first during setup and then `singleton_mtx`. However during destruction we acquire `singleton_mtx` first and then `remote_elem_mtx`. This has the potential to deadlock. So why not let the singleton code handle all the locking? Seems like singleton code should do that kind of thing and we should leverage it in `RemoteElem` if we are going to go through the process of inheriting from `Singleton`. This reports helgrind clean now
I ran across this speedbump while trying out some alternate configs, which I needed for debugging the quad-precision support additions I made recently, which I did to help debugging the newest IGA code. The problem has now been whipped. We often have holes in our id numbering in the distributed mesh case, so to index by id we need vectors that can hold the maximum id number of an element, not just the number of elements. Anyone who's reading this commit message has decided to inspect not flee the tale of the flog of the bump in our logs on the hole in the bottom of the C.
If we run our unit tests with --re '^A' then we really want just *tests* that start with A, not "All Tests", even though All starts with A.
There's a chance this is responsible for a valgrind warning that @fdkong managed to trigger.
…lerances Otherwise, we won't use custom tolerances when doing the tree search, so we might end up doing way more linear searches than is necessary.
After updating these metrics in 60487b5, I should have also updated the bounds. Note that Knupp does not give explicit bounds on these metrics for a "good" quality element in his paper, but both have an upper bound of 1, so the previous values for SKEW were just wrong.
The user will now have to opt-in to gdb backtraces instead of opting out with `--without-gdb-command`. There are many times that I forget to configure with `--without-gdb-command` and then I sadly wait for a minute to get my prompt back or switch to another terminal to kill the process. For all of MOOSE testing we run with `--no-gdb-backtrace`. In general if I need a backtrace I will compile in debug mode and print a backtrace. I almost never make use of the automatic backtrace output; instead I just dislike it. There was some general support for this change on our libmesh slack channel
|
Nope, "Continuous integration has not been set up". |
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
Trying to run our Civet tests of 1.6.1.
Not sure if I can get moosebuild to play with this PR, since the base repo is in roystgnr/libmesh and not libMesh/libmesh, but let's try things this way first before I pollute the libMesh/ branch namespace.