We can now properly suppress the ubsan warnings about
_Sp_counted_base.

Fix stacktrace_test for the case where it is compiled using a relative,
not absolute, path to the source.

LockedPointer template class, to allow returning a held lock along
with a pointer in a movable object.

Add ATH_MEMBER_REQUIRES/ATH_MEMBER_REQUIRES_DEF to concepts.h.

gcc seems to support __builtin_convertvector already in version 9. We should be compiling master with clang>=10 anyhow so no need to check for clang support for the ternary operator for vectorized types

Leftover debugging pragma.

With newer lcg versions, the system addr2line doesn't work because it's too old.
Normally that doesn't matter, because we usually use eu-addr2line, which does
work.  However, the stacktrace unit test forces the use of /usr/bin/addr2line
(as the output formats are slightly different).

Change stacktrace so that if we don't find /usr/bin/eu-addr2line we
search PATH for addr2line, rather than hardcoding it.  Also adjust the
unit test so that it should work with either addr2line or eu-addr2line.

Accessing a thread_local variable may result in a call to malloc.
Avoid using thread_local in code that runs as part of the fast stack dump,
so that the dump will work even if the heap is corrupt.

See ATR-22583.

Fix clang12 warnings about uninitialized variables.

Newer glibc versions change some details in how signals are handled.
Update the stacktrace test to not depend on this.

Add a feature test macro for mallinfo2().  (New in glibc 2.33.)

Add FPControl helper, to allow temporarily disabling FPEs.

Attila Krasznahorkay authored 3 years ago and Frank Winklmeier committed 3 years ago

The "linux implementation" of the code is only appropriate for x86, so the code
will be compiled like that from now on. Plus fixed a small issue in the
"dummy implementation" that prevented it from working with the rest of the code.

Suppress cppcheck warnings.

Add code to make a nice hex dump of memory, optionally with protection
against trying to read unmapped regions of memory.

Change procmaps to use a binary search rather than a linear search.
In passing, fixes a bug where it returned the wrong range when the
address being searched for was at the start of a range.

Ewelina Maria Lobodzinska authored 3 years ago and Atlas Nightlybuild committed 3 years ago

Making sure MGC cleans up copies of events

See merge request !42973

(cherry picked from commit 3e3831a3)

ece0bc68 Making sure MGC cleans up copies of events

Ewelina Maria Lobodzinska authored 3 years ago and Atlas Nightlybuild committed 3 years ago

MGC tests, saving tgz outputs instead of event files

See merge request !42977

(cherry picked from commit 3bf41984)

5fab50b6 MGC tests, saving tgz outputs instead of event files

Nils Erik Krumnack authored 3 years ago and Atlas Nightlybuild committed 3 years ago

DL2 umami

See merge request !42873

(cherry picked from commit c5e0e712)

731fbc6f DL2 adaptions to be able to run umami

John Derek Chapman authored 3 years ago and Atlas Nightlybuild committed 3 years ago

Fix the mm cross talk implementation to handle cross talk 0

See merge request !42750

(cherry picked from commit 059cfabc)

7b7f3470 fix of cross talk simulation to work with 0 cross talk in mm digi
e0f544ff fix whitespace

Oana Vickey Boeriu authored 3 years ago and Atlas Nightlybuild committed 3 years ago

Fix issue with short index for IOVPayload PT conversion

See merge request !42384

(cherry picked from commit 43f733e8)

43b3fded Fix issue with short index for IOVPayload PT conversion

John Derek Chapman authored 3 years ago and Atlas Nightlybuild committed 3 years ago

cleanup of MMClusterOnTrack

See merge request !42298

(cherry picked from commit 55c1083c)

13f1eebd cleanup of MMClusterOnTrack

Fix signature of __cxa_throw.

Update the exctrace collector to work properly in a MT environment.
Change the static variables holding the trace to be thread_local.
Further, retrieve the saved trace via calling a function, rather
than accessing the variables directly.

std:is_pod is deprecated in c++20.

If a class is declared with a requires clause, the equivalent clause
must be present for all function definitions outside of the class body.

Suppress cppcheck false positives.

Suppress cppcheck false positive about missing dtor for ConcurrentHashmapImpl.
m_table gets deleted because it's handed to the Updator object.

Fix a couple typos in comments.

backtrace() will stop if it hits an invalid stack frame,
which can make it impossible to diagnose some kinds of crashes.

Try to hack it so that in many cases it will be able to proceed
past one bad frame.

See ATR-22773.

The Intel compiler is no longer using the macro name CL_SYCL_LANGUAGE_VERSION,
but rather SYCL_LANGUAGE_VERSION.

At the same time still kept the CL_SYCL_LANGUAGE_VERSION check around as
well, in case we would try to build the code with an older version of the
Intel compiler.

New compiler, new set of warnings...

The main theme here is that clang now checks for range-based for loops
whether it introduces unneeded temporaries, i.e. if the user asks for
a reference for the loop variable, but the iterator doesn't return a
reference or a reference of a different type, or conversely if the
user asked for `const auto`, but could have been `const auto&`.

In practice that usually hits us as:
```
for (const auto& jet : jetContainer)
```
which should be:
```
for (const auto *jet : jetContainer)
```

Also a couple of places like this (which miss the `const` qualifier on
the first template parameter):
```
std::map<std::string,std::string> stringMap;
for (const std::pair<std::string,std::string>& stringPair : stringMap)
```

For the most part I just replaced the loop variable type with what
seemed correct.  In a few places I put explicit comments as to why I
chose the type.

Also a fair number of warnings for unused member variables in various
packages.  Since I am not an expert on any of these packages and this
can point to an actual bug, I commented out all unused variables and
added a comment for actual experts to check and remove.

I removed a couple of checks for `this != nullptr` which were
originally introduced to check whether the user called a member
function via a null pointer.  However those checks are `assert`-based,
so with cmake they won't be included when the user calls them from a
release, making this check mostly useless.

There some other warnings I fixed in the process that should hopefully
be self-explanatory.

Fix a typo in comments.

It's non-ATLAS, public-domain code.
Remove ATLAS copyright; restore original copyright notice that
had been mangled.