bug fix, generated error with %%F or %%G in PetscPrintf output

completed removal of %G from PETSc, we now generate an error if used

fixes for quad precision

for some reason %G sometimes works and sometimes doesn't needs debugging

Sys: drop explicit type arguments from PetscNew() and PetscNewLog()

git grep -l 'PetscNew\(Log\)\?(' -- $1 | xargs perl -pi -e '
s@PetscNew\([^,;()]+ *, *@PetscNew(@;
s@PetscNewLog\(([^,;()]+) *,[^,

Sys: drop explicit type arguments from PetscNew() and PetscNewLog()

git grep -l 'PetscNew\(Log\)\?(' -- $1 | xargs perl -pi -e '
s@PetscNew\([^,;()]+ *, *@PetscNew(@;
s@PetscNewLog\(([^,;()]+) *,[^,;()]+, *@PetscNewLog($1,@'

show more ...

Sys: add PetscMalloc1 macro, array allocation without redundant types

The type is inferred from the pointer return type. This patch is
automated via the following script:

git grep -l 'PetscMalloc(

Sys: add PetscMalloc1 macro, array allocation without redundant types

The type is inferred from the pointer return type. This patch is
automated via the following script:

git grep -l 'PetscMalloc(.*sizeof' src | xargs perl -pi -e 's@PetscMalloc\(([^,;]*[^,; ]) *\* *sizeof\([^,;()]+\),@PetscMalloc1($1,@'

This commit contains an additional bug-fix in csrperm.c, fixing pointer
arity. The code was introduced in 2006, but the allocation could not
have been correct at any time. This probably means that
MatDuplicate_SeqAIJPERM has never been tested.

a54129beb540034ba105796c682d589e7e1111f2
Richard Tran Mills <rmills@ornl.gov>

Added MATSEQCSRPERM support for MatDuplicate() and conversion to/from
MATSEQAIJ. Note that these changes are not quite debugged.

show more ...

added FILE* argument to PetscSynchronizedFlush() so it knows where to flush the off process values

Previously if process 0 did not call PetscSynchronizedFPrintf() then it output the off process valu

added FILE* argument to PetscSynchronizedFlush() so it knows where to flush the off process values

Previously if process 0 did not call PetscSynchronizedFPrintf() then it output the off process values
to stdout even if the file was not stdout. Also affected PetscViewerASCIISynchronizeFlush()

Reported-by: Randall Mackie <rlmackie862@gmail.com>

show more ...

remove non-c89 initialization of structs
update serialization of functions code

Merge branch 'jed/fix-synchronized-fgets'

Fixed deadlock when PetscSynchronizedFGets hits EOF since MPI_Bcast size
is mismatched.

* jed/fix-synchronized-fgets:
PetscSynchronizedFGets: fix deadloc

Merge branch 'jed/fix-synchronized-fgets'

Fixed deadlock when PetscSynchronizedFGets hits EOF since MPI_Bcast size
is mismatched.

* jed/fix-synchronized-fgets:
PetscSynchronizedFGets: fix deadlock at EOF
Compiler: Fix warnings from MPI impls which do not initialize outputs

show more ...

PetscSynchronizedFGets: fix deadlock at EOF

> Problem: When reading a file of unknown length using
> PetscSynchronizedFGets(), all but the first process in the communicator
> freeze after reaching t

PetscSynchronizedFGets: fix deadlock at EOF

> Problem: When reading a file of unknown length using
> PetscSynchronizedFGets(), all but the first process in the communicator
> freeze after reaching the end of the file.
>
> Cause: When EOF is reached, PetscSynchronizedFGets() sets len = 0.
> However, this only happens in the first process, so the other processes
> are still expecting the originally requested number of bytes to be
> broadcast to them, and thus hang at MPI_Bcast().
>
> Solution: Do not set len = 0 at EOF.

This commit cherry-picks relevant portions of multiple commits from
the feature branch 'barry/wirth-fusion-materials'.

Reported-by: Ilmari Karonen <ilmari.karonen@helsinki.fi>

show more ...

Compiler: Fix warnings from MPI impls which do not initialize outputs

(cherry picked from commit d892089bd4a83d5450f25ac15dd2b0a661537089)

Compiler: Fix warnings from MPI impls which do not initialize outputs

Merge branch 'jed/avoid-getpwuid'

The file path routines were broken because the current directory was
placed in front of the given path before checking for leading ~ or
~user. This is now fixed, al

Merge branch 'jed/avoid-getpwuid'

The file path routines were broken because the current directory was
placed in front of the given path before checking for leading ~ or
~user. This is now fixed, along with the removal of getpwuid (a
liability on BG/Q and Cray). We still attempt to use getpwnam when
encountering ~user.

* jed/avoid-getpwuid:
Sys ex13: avoid non-portable realpath(1) in verification test
cleanup includes that are no longer needed. Also remove check for _fullpath() as the corresponding code is removed.
Sys: fix PetscGetFullPath expansion of ~ and ~user, ensure NULL termination
Sys: avoid getpwuid in file path routines
Sys: add failing test expanding ~/file-name and ~user/file-name

Conflicts:
config/PETSc/Configure.py

show more ...

cleanup includes that are no longer needed. Also remove check for _fullpath() as
the corresponding code is removed.

Sys: fix PetscGetFullPath expansion of ~ and ~user, ensure NULL termination

Also update test to check that paths are handled as well as possible
when the paths are too short. Still not bulletproof.

Sys: fix PetscGetFullPath expansion of ~ and ~user, ensure NULL termination

Also update test to check that paths are handled as well as possible
when the paths are too short. Still not bulletproof.

If the path is much too short (not tested here), then
PetscGetWorkingDirectory will fail with an error because getcwd fails.

show more ...

Sys: avoid getpwuid in file path routines

This function is badly implemented by some well-known supercomputing
vendors (causes corruption and crashes) and ~ expansion in the shell
uses $HOME rather

Sys: avoid getpwuid in file path routines

This function is badly implemented by some well-known supercomputing
vendors (causes corruption and crashes) and ~ expansion in the shell
uses $HOME rather than the password home directory field anyway.

This commit does not affect the failing test for path expansion.

show more ...