The fix in the DeviceManager is more like a work-around for a false
positive of gcc 13.

Dietrich Rothe authored 7 months ago and Dietrich Rothe committed 7 months ago

* most useful with StatusOutputs, where automatic propagation of
  'data invalid' status is often not wanted
* Change default behavior of StatusOutput and StatusWithMessage.
  By default, they no longer propagate DataValidity=invalid of owning module.
  They still propage DataValidity=invalid of explicitly set for the output.
* Adapt test of DataValidity propagation for StatusOutput
* Adapt test of StatusMonitor to new StatusOutput behaviour

Dietrich Rothe authored 7 months ago and Dietrich Rothe committed 7 months ago

"!sometag" is a negated tag which annihilates with "sometag"
negated tags can be added like normal tags, and are used to remove their counterparts

boost::thread::id seems not to be trivially copyable for the old boost
version, so we implement our own atomic access with a lock here.

Dietrich Rothe authored 9 months ago and Dietrich Rothe committed 9 months ago

CircularDependencyDetector was running into infinite recursion (causing
stack overflow) in a situation of two Modules which are both waiting on
each other for initial values

* fix: Ignore data validity through return channels

* chore: Minor linter fixes

* fix: Do not modify model while iterating in XML generator

* fix: Change XML generator for TriggerReceivers

Give it a chance to comply with the XSD we provide

* fix: Update XML output XSD for changed output

 - Add Void and Boolean
 - Add TriggerReciever
 - Make class optional

* chore: Update test for XML generator

Add a DeviceModule whose absence was causing the XML generator to fail unnoticed

Modifying the model while visiting is not allowed because the
modification invalidates the iterators used for visiting. This lead in
this case to a very subtle crash (segfault) only appearing under very
specific circumstances (only when running in ctest and only on some
Ubuntu 20.04 machines).

ProcessVariableProxy::removeNode() was effectively taking a reference to
a shared pointer as an argument and is removing the same (effective)
shared pointer from a list. This poses a problem if the reference passed
points to the list directly and is the last shared pointer to this
object so it gets freed, because removeNode() is accessing the object
after removing it from the list.

BREAKING CHANGE:
Since previously more characters were allowed (with possibly adverse
effects), this commit may force application developers and/or system
integrators to take action by changing variable names such that they
comply to the rules. Only alphanumeric characters and the underscore
are allowed, while slashes (and dots) are understood as hierarchy
separators.

- fix EntityOwner::constant() violates character restrictions
- Utilities::stripName() was broken in multiple ways. It has been
  changed into a true, reversible escaping algorithm and subsequently
  renamed into Utilities::escapeName().
- Utilities::unescapeName() is introduced.

BREAKING CHANGE: This commit will break applications which either
- rely on the names of device status PVs for devices which contain
  illegal characters (e.g. when using a CDD directly instead of a
  dmap file alias), or
- use Utilities::stripName() directly.

* Disable deprecation warnings
* Fix linter findings
* fix: Compiler warnings regarding internal linkage
* fix: Mask memory leak

---------

Co-authored-by: Martin Hierholzer <martin.hierholzer@desy.de>

InvalidityTracer and ConnectionMaker debug mode print now information
about circular networks

Constants (as generated by ApplicationModule::constant()) were causing a
logic_error if accessors of different types were using the same constant
(created with the same value and the same value type upon creation).
This is possible since the type passed to constant() does not have to be
the same as the type of the accessor.

This is now fixed by giving each constant a fully unique name, so every
accessor using a constant will be in its own little variable network.

This is the implementation for the specification change done in a recent
commit.

May occur when not mapping a Void variable in the control system so it
is being optimised out.