What’s new?

What’s new?#

The sections below will give you the most relevant news from the PyTango releases. For help moving to a new release, or for the complete list of changes, see the following links:

Unreleased#

Type: in development

Here we provide highlights of the changes currently on the development branch.

Changed#

  • tango.server.device_property values are now initialised from same-named class properties when no device database value is defined, matching the precedence of C++ device servers generated with POGO. Values are resolved in this order: the device database value, the class database value, the declared class property default, and the declared device property default. For example, given this device:

    class PowerSupply(Device):
  port = device_property(dtype=int, default_value=5555)

    • if the Tango database has a property set at device level, that value is used;

    • otherwise, if it is set at class level, the class database value is used;

    • otherwise, the device property default is used: 5555 in this example.

    • Note: with the high-level API we can’t declare port as class_property as well.

    So, if you have a property set at class level in your Tango database that matches the name of a device property, and you don’t have the device property set in the Tango database, then the behaviour of that device will change:

    • it will now use the class-level database property instead of the code default,

    • a same-named class property default will take precedence over the device property default,

    • if the device property is mandatory, the device will now start, using the class-level value, instead of raising an exception.

  • Removed deprecation warnings when old telemetry environment variable names are set. They remain ignored for configuration. This allows different versions of PyTango to be run in the same environment.

  • Removed automatic refresh of client telemetry configuration, since this could not actually have an effect without restart. Client telemetry configuration is only captured from environment variables and fallback configuration files when Tango is imported.

Added#

  • The read, write and is_allowed handlers of a tango.server.attribute can now be called from within device code through the tango.Attribute or tango.WAttribute object: self.my_attr.read(), self.my_attr.write(value) and self.my_attr.is_allowed(request_type). The same direct access is available on any tango.Attribute object of that attribute, however it was obtained, for static as well as for dynamic high-level attributes. Classic API attributes, and dynamic attributes added as low-level Attr/AttrData objects, raise NotImplementedError. The write method is available only on writable tango.WAttribute objects. The handlers run with the user’s own decorators applied (e.g. DebugIt). This is a shortcut allowing a direct call to your method, so Tango’s internal mechanisms that run when remote clients access your attributes: i.e., always_executed_hook, read_attr_hardware, is_allowed handlers and telemetry are not triggered. Your handler is called exactly as you wrote it, so the call returns an awaitable only if your handler is a coroutine function. When the attribute has no is_allowed handler, is_allowed answers True in the green mode of your device. As for any other tango.Attribute method, do not keep a reference to the attribute object across a remove_attribute call: the C++ object behind it is destroyed with the attribute.

  • tango.Attribute.get_att_device(), which returns the device the attribute belongs to.

Fixed#

  • Fixed many deadlocks and stalled Python threads during event cleanup and queue access, callback and group operations, connections, database and attribute property calls, server polling and lifecycle operations, and runtime attribute configuration. These calls now release the GIL during blocking cppTango work, and reacquire the GIL before Python cleanup.

  • tango.Database.get_property_forced() no longer raises a TypeError. The binding required the trailing DbServerCache argument of the C++ call, but that class is not bound, so the argument could not be supplied and has now been dropped.

  • tango.Database.get_device_property_list() no longer raises a TypeError when called with a tango.StdStringVector container, for the same reason, and now returns that container instead of None.

  • tango.Database.build_connection() now calls the matching cppTango method instead of writing the file database.

  • Elements of a DevVarCharArray are now range-checked: values outside the 0-255 range no longer wrap silently to 255 and 0, so 256 raises an OverflowError and -1 a TypeError. A non-integer element is now reported as “Expecting a integer type”, instead of “Expecting a char type”.

  • Unsigned integer types now accept any object implementing __index__, numpy integer scalars of any width included, instead of only python integers and numpy scalars of the exactly matching dtype. This applies to DevUChar, DevUShort, DevULong and DevULong64, both scalar and array.

  • The tango.Except.print_exception(), which is listed in the Except docstring but was forgotten to be added in bindings during move to pybind11, is now added back. tango.device_class calls it when a tango.DeviceClass constructor raises, so the original DevFailed was replaced by an AttributeError.

  • tango.DeviceProxy.write_attribute() no longer crashes or hangs when the attribute does not exist. The failure of the attribute configuration lookup was swallowed, so the value was converted against an uninitialised configuration: on macOS this segfaulted, on Linux it raised “unsupported data_format”. The DevFailed is now propagated, matching write_attributes(), and an attribute data type that the conversion cannot handle now raises a TypeError naming it instead of being silently ignored in release builds.

What’s new in PyTango 10.3.1?#

Date: 2026-07-20

Type: patch release

Changed#

  • Calling the tango.Util.cleanup() method from the core server Tango loop is now optional (default=False). The call was added unconditionally in 10.3.0 to be able to restart the server loop in the same process. However, since this feature was only added to cppTango experimentally, it has introduced some bugs. For example, 10.3.0 devices servers would often segfault at the end of a normal shut down. For the (Multi)DeviceTestContext the default is set to True, so you can still create multiple test contexts in the same python process.

Added#

Fixed#

  • Memory leak when Python exceptions are converted to DevFailed exceptions. This was most noticeable if Python exceptions were raised in attribute handlers read very often. The bug was a regression introduced in 10.1.0.

  • PyTango now supports NumPy 2.5

Packaging#

Starting from this release:

  • PyTango no longer provides compiled wheels for the i686 platform, following NumPy’s example.

  • The Linux builder has moved from the manylinux2014 image to the manylinux_2_28 image (following NumPy 2.5). This means the Linux wheels can only be installed on Linux platforms with at least glibc 2.28 (including Debian 10+, Ubuntu 18.10+, Fedora 29+, CentOS/RHEL 8+).

  • The macOS image was updated to macos-15-xcode-16.

General:

  • PyTango packages are compiled with support for OpenTelemetry in Linux and macOS, but not Windows:

Compilation option

Linux

Windows

MacOS

TANGO_USE_TELEMETRY

on

off

on

  • Some of the dependencies packaged with the binary wheels on PyPI have changed. The bundled versions are:

Dependency

Linux

Windows

MacOS

cpptango

10.3.3

10.3.3

10.3.3

omniorb / omniorb-libs

4.3.4

4.3.3

4.3.4

libzmq / zeromq

4.3.5

4.0.5-2

4.3.5

cppzmq

4.11.0

4.7.1

4.11.0

libjpeg-turbo

3.1.90

2.0.3

3.2.0

abseil

20260107.0

-

20260526.0

protobuf

v33.6

-

v35.1

c-ares

v1.34.6

-

1.34.8

re2

2025-11-05

-

2025.11.05

OpenSSL

3.6.2

-

3.6.3

curl

curl-8_20_0

-

8.21.0

gRPC

v1.80.0

-

1.82.1

opentelemetry-cpp

v1.27.0

-

1.27.0

What’s new in PyTango 10.3.0?#

Date: 2026-06-03

Type: major release

Here we provide highlights of the changes currently on the development branch.

Changed#

  • PyTango requires at least cppTango 10.3.0. See the migration guide.

  • The telemetry environment variables now match cppTango 10.3.0, using the renamed TANGO_TELEMETRY_TRACING_* and TANGO_TELEMETRY_LOGGING_* variables, and warning if the old names are used. The old names are no longer used for configuration. See the telemetry how-to guide.

  • Device telemetry can now be reconfigured at runtime instead of being treated as an import-time-only setup on the Python side. PyTango recreates or replaces its Python tracer providers when cppTango runtime configuration changes. See the telemetry runtime configuration how-to. Note that for pure client scripts, the telemetry configuration will still be based on the environment variables at import time.

  • Telemetry topics can be used, but the only useful option at the moment is all. The user topic can result is dis-jointed or incomplete traces. See telemetry topics.

  • Missing OpenTelemetry Python dependencies now warn only when telemetry is actually requested, instead of warning merely at import time. See the telemetry how-to guide.

Added#

Fixed#

  • Fixed a regression (introduced in PyTango 10.1.0) where attribute event threshold properties (rel_* / abs_*) were silently ignored when subscribing to change or archive events, causing subscriptions to succeed even when required thresholds were not configured.

  • write_attribute_asynch() now reports the full error stack on failure. Previously, internal delegation to write_attributes_asynch produced incomplete error output.

  • register_signal() can now be called without the optional own_handler argument. Previously omitting it raised a TypeError.

  • Fixed a deadlock in asyncio device servers where cppTango device functions were called while still holding the GIL. This manifested as communication timeouts when asyncio servers were monitored by Jive ATK panels (see issue #629).

  • Fixed a memory leak on Python 3.14+ where DeviceAttribute values and many other objects were never freed due to a pybind11 bug in managed-dict handling. 19 classes are affected, including DeviceProxy, Database, and all EventData types.

  • Fixed subscribe_event failing when called from within an asyncio device server (self-subscription). Thread-pool threads could not locate the correct green executor, preventing event subscriptions from being registered (see issue #747).

  • Fixed a crash when constructing Database objects caused by a pybind11 limitation with GIL-release call-guards on constructors.

  • Fixed silent coercion of typos (e.g. "ture", "yse") to False when parsing string device property values as booleans. String-to-boolean conversion now strictly accepts only y/yes/t/true/on/1 and n/no/f/false/off/0 (case-insensitive), and raises an error otherwise. Errors during property parsing also now report the device name, property name, raw value, and target type to help diagnose configuration issues.

  • Pulled in upstream changes from pytango-db 0.6.0, updating the bundled Python DatabaseDS implementation.

Removed#

  • Removed the tango.server.Device.set_kernel_tracing_enabled() and tango.server.Device.is_kernel_tracing_enabled() methods. These methods should not have been available within a device. Instead, clients can enable kernel tracing by setting the telemetry topic "all".

Packaging#

  • PyTango packages are compiled with support for OpenTelemetry in Linux and macOS, but not Windows:

Compilation option

Linux

Windows

MacOS

TANGO_USE_TELEMETRY

on

off

on

  • Some of the dependencies packaged with the binary wheels on PyPI have changed. The bundled versions are:

Dependency

Linux

Windows

MacOS

cpptango

10.3.2

10.3.2

10.3.2

omniorb / omniorb-libs

4.3.4

4.3.3

4.3.4

libzmq / zeromq

4.3.5

4.0.5-2

4.3.5

cppzmq

4.11.0

4.7.1

4.11.0

libjpeg-turbo

3.1.90

2.0.3

3.1.4.1

abseil

20260107.1

-

20260107.1

protobuf

v33.6

-

v33.5

c-ares

v1.34.6

-

1.34.6

re2

2025-11-05

-

2025.11.05

OpenSSL

3.6.2

-

3.6.2

curl

curl-8_20_0

-

8.20.0

gRPC

v1.80.0

-

1.78.1

opentelemetry-cpp

v1.27.0

-

1.27.0

What’s new in PyTango 10.1.6?#

Date: 2026-09-04

Type: patch release

Fixed#

  • Fixed a deadlock in asyncio device servers where cppTango device functions were called while still holding the GIL (backport from 10.3.0).

  • Fixed a possible deadlock with AttributeProxy and Database during object destruction (regression since 10.1.0, backport from planned 10.3.2).

  • Fixed a crash when constructing Database objects caused by a pybind11 limitation with GIL-release call guards on constructors (backport from 10.3.0).

  • Fixed numpy.typing import error with NumPy < 2 (regression since 10.1.5, backport from planned 10.3.2)

  • Fixed several rare memory leaks and potential crashes in the C++ bindings (backport from planned 10.3.2).

  • Corrected ownership transfer for nested Group objects, preventing memory leaks, and possible segmentation faults when groups are attached to a parent (backport from planned 10.3.2).

Packaging#

General:

  • PyTango packages are compiled with support for OpenTelemetry in Linux and macOS, but not Windows:

Compilation option

Linux

Windows

MacOS

TANGO_USE_TELEMETRY

on

off

on

  • Some of the dependencies packaged with the binary wheels on PyPI have changed. The bundled versions are:

Dependency

Linux

Windows

MacOS

cpptango

10.1.2

10.1.2

10.1.2

omniorb / omniorb-libs

4.3.4

4.3.0

4.3.4

libzmq / zeromq

4.3.5

4.0.5-2

4.3.5

cppzmq

4.11.0

4.7.1

4.11.0

libjpeg-turbo

3.2.0

2.0.3

3.2.0

abseil

20260526.0

-

20260107.1

protobuf

v35.1

-

33.5

c-ares

v1.34.8

-

1.34.8

re2

2025-11-05

-

2025.11.05

OpenSSL

3.6.3

-

3.6.4

curl

curl-8_21_0

-

8.22.0

gRPC

v1.82.1

-

1.78.1

opentelemetry-cpp

v1.28.0

-

1.21.0

What’s new in PyTango 10.1.5?#

Date: 2026-07-21

Type: patch release

Added#

Fixed#

  • Memory leak when Python exceptions are converted to DevFailed exceptions (backport from 10.3.1). This was most noticeable if Python exceptions were raised in attribute handlers read very often. The bug was a regression introduced in 10.1.0.

  • Support of NumPy 2.5 (backport from 10.3.1)

Packaging#

Similar to 10.3.1, in this release:

  • PyTango no longer provides compiled wheels for the i686 platform, following NumPy’s example.

  • The Linux builder has moved from the manylinux2014 image to the manylinux_2_28 image (following NumPy 2.5). This means the Linux wheels can only be installed on Linux platforms with at least glibc 2.28 (including Debian 10+, Ubuntu 18.10+, Fedora 29+, CentOS/RHEL 8+).

  • The macOS image was updated to macos-15-xcode-16.

General:

  • PyTango packages are compiled with support for OpenTelemetry in Linux and macOS, but not Windows:

Compilation option

Linux

Windows

MacOS

TANGO_USE_TELEMETRY

on

off

on

  • Some of the dependencies packaged with the binary wheels on PyPI have changed. The bundled versions are:

Dependency

Linux

Windows

MacOS

cpptango

10.1.2

10.1.2

10.1.2

omniorb / omniorb-libs

4.3.4

4.3.0

4.3.4

libzmq / zeromq

4.3.5

4.0.5-2

4.3.5

cppzmq

4.11.0

4.7.1

4.11.0

libjpeg-turbo

3.2.0

2.0.3

3.2.0

abseil

20260526.0

-

20260107.1

protobuf

v35.1

-

v33.5

c-ares

v1.34.8

-

1.34.8

re2

2025-11-05

-

2025.11.05

OpenSSL

3.6.3

-

3.6.3

curl

curl-8_21_0

-

8.21.0

gRPC

v1.82.1

-

1.78.1

opentelemetry-cpp

v1.28.0

-

1.21.0

What’s new in PyTango 10.1.4?#

Date: 2026-03-03

Type: patch release

Fixed#

  • Fixed pybind11 managed dict leak for all py::dynamic_attr() classes on Python 3.14+. This leak, in particular, was seen in DeviceProxy and AttributeProxy when reading attributes and receiving events with Python 3.14. But several other classes were also affected. See issue #744 and MR !980.

The packing details are the same as 10.1.3.

What’s new in PyTango 10.1.3?#

Date: 2026-02-10

Type: patch release

This release contains no changes in PyTango, only new packing details for the wheels:

Packaging#

  • PyTango packages are compiled with support for OpenTelemetry in Linux and macOS, but not Windows:

Compilation option

Linux

Windows

MacOS

TANGO_USE_TELEMETRY

on

off

on

  • Some of the dependencies packaged with the binary wheels on PyPI have changed. The bundled versions are:

Dependency

Linux

Windows

MacOS

cpptango

10.1.2

10.1.2

10.1.2

omniorb / omniorb-libs

4.3.4

4.3.0

4.3.4

libzmq / zeromq

4.3.5

4.0.5-2

4.3.5

cppzmq

4.11.0

4.7.1

4.11.0

libjpeg-turbo

3.1.3

2.0.3

3.1.2

abseil

20250814.1

-

20250512.1

protobuf

v33.5

-

v31.1

c-ares

v1.34.6

-

1.34.6

re2

2025-11-05

-

2025.11.05

OpenSSL

3.6.1

-

3.6.1

curl

curl-8_18_0

-

8.18.0

gRPC

v1.76.0

-

1.73.1

opentelemetry-cpp

v1.24.0

-

1.21.0

What’s new in PyTango 10.1.2?#

Date: 2026-01-09

Type: patch release

Fixed#

  • Regressions in 10.1.2:

    • when subscribing to change/archive events properties are ignored

    • DevVoid commands mutate type if decorator changes the type of the argument.

The packing details are the same as 10.1.1.

What’s new in PyTango 10.1.1?#

Date: 2025-10-31

Type: patch release

Fixed#

  • Regression in 10.1.0 of image (2-D) attributes under Windows, 64-bit.

The packing details are the same as 10.1.0.

What’s new in PyTango 10.1.0?#

Date: 2025-10-28

Type: major release

Here we provide highlights of the changes:

(You can find the full changelog in the history of changes with the links to related MRs, where the detailed descriptions of every change/new feature can be found.)

Changed#

  • PyTango requires at least cppTango 10.1.0. See the migration guide.

  • The biggest change in this release is moving the C++ Python bindings from Boost.Python to Pybind11. There are a few breaking changes. See the migration guide.

  • The parameter names for attribute write methods, such as write_attribute, write_attributes, write_read_attribute, write_read_attributes, write_attribute_async write_attributes_async for DeviceProxy, AttributeProxy and Group were changed to be consistent. See migration guide.

  • When subscribing to events, there are now more options, including asynchronous mode. This new mode can significantly speed up subscription to a large number of attributes. See the migration guide.

  • The string representation of nested structs has been improved so they print a little more nicely.

  • References to event subscription callback methods are immediately cleared on unsubscription (instead of sticking around for at least a minute).

  • Clients (DeviceProxy, AttributeProxy, Group) and servers now release the Python GIL in most cases when waiting on cppTango, allowing better concurrency, and avoiding some deadlocks. Note: The Database client wasn’t updated, and still holds the GIL for most calls.

  • Better types hints for the extension code (in stub files), and the stub files are now included in the conda-forge osx-arm64 packages.

  • Many improvements to the documentation, including for attribute.

  • The MultiDeviceTestContext and DeviceTestContext now check if the device server could be properly shutdown (when exiting the context handler, or calling stop). If the device server could not be stopped, a RuntimeError is raised. This will help identify problematic device servers during testing. If using a subprocess (process=True), the subprocess will be killed.

  • Support Coverage.py runs using sys.monitoring (aka the “sysmon” core).

  • Removed limitation of NumPy version < 2.3 at build time.

  • Signatures of put_property, get_property and delete_property methods of DeviceProxy and AttributeProxy objects were unified. get_property method now accepts dicts and in case of empty list/tuple/dict input returns an empty dict (before it could return None in some cases)

  • For the Python-based DatabaseDS server (developer tool):

    • Some of the “default” values in the database have been improved to fix a problem with getting device info for the DB device. However, it will only take effect when creating a new database from scratch. If you don’t want to do that you can also fix them by running a command like this:

      • sqlite3 tango_database.db "UPDATE device SET pid='0' WHERE pid = 'nada';"

    • SQLite databases created using earlier versions of PyTango might have issues with some commands not working. E.g., at startup you could get “device dserver/databaseds/2 not defined in the database !”. If that happens, create new database file from scratch. If you need to keep your data, you could use a tool like dsconfig to dump the old database to JSON and load it into the new database.

Added#

  • Python 3.14 support.

  • When defining attributes, you can indicate that they push events in place, removing the need for extra calls to set_change_event, etc. See the migration guide.

  • Command handler docstrings can now populate the in_type_desc and out_type_desc fields automatically. There is support for ReST, Google, Numpydoc-style and Epydoc docstrings. See the migration guide.

  • The client identity is now available in devices when handling command or attribute read/write calls. However, only for synchronous green mode device servers. See tango.server.Device.get_client_ident(). If OpenTelemetry is enabled, the client details will be included in relevant spans (for all green modes).

  • Non-synchronous green mode clients now keep OpenTelemetry context for all traced calls.

  • Two test utility functions: tango.test_utils.wait_for_proxy() and tango.test_utils.wait_for_nodb_proxy_via_pid().

  • Added fast overload for attribute write methods, such as write_attribute, write_attributes, write_read_attribute, write_read_attributes, write_attribute_async write_attributes_async, write, write_read, write_asynch for DeviceProxy, AttributeProxy and Group with which the additional IO to server can be avoided. See migration guide.

  • Default values for device_property and class_property can be also provided as the string representation (array of string representations) of the value(s)

Fixed#

  • Dynamic commands with no argument (regression in 10.0.0).

  • pre_init_callback and post_init_callback used with a sequence (regression in 10.0.0).

  • Type hints with decorated commands - order no longer matters.

  • Type hints are parsed when from __future__ import annotations is used.

  • Crash when exception raised in delete_device.

  • AttributeProxy’s slow initialization (remove unnecessary listing of all commands and attributes).

Removed#

  • Support for Python 3.9.

  • Support for Tango pipes.

  • The dim_x and dim_y parameters for setting attribute values and pushing events. See the migration guide. The size is automatically determined from the data.

Packaging#

  • PyTango packages are compiled with support for OpenTelemetry in Linux and macOS, but not Windows:

Compilation option

Linux

Windows

MacOS

TANGO_USE_TELEMETRY

on

off

on

  • Some of the dependencies packaged with the binary wheels on PyPI have changed. The bundled versions are:

Dependency

Linux

Windows

MacOS

cpptango

10.1.1

10.1.1

10.1.1

omniorb / omniorb-libs

4.3.3

4.3.0

4.3.3

libzmq / zeromq

4.3.5

4.0.5-2

4.3.5

cppzmq

4.11.0

4.7.1

4.10.0

libjpeg-turbo

3.1.2

2.0.3

3.1.0

abseil

20250814.1

-

20250512.1

protobuf

v32.1

-

v31.1

c-ares

v1.34.5

-

1.34.5

re2

2025-08-12

-

2025.08.12

OpenSSL

3.6.0

-

3.5.4

curl

curl-8_16_0

-

8.16.0

gRPC

v1.75.1

-

1.73.1

opentelemetry-cpp

v1.23.0

-

1.21.0

What’s new in PyTango 10.0.4?#

Date: 2026-02-10

Type: patch release

This release contains no changes in PyTango, only new packing details for the wheels:

Packaging#

  • PyTango packages are compiled with support for OpenTelemetry in Linux and macOS, but not Windows:

Compilation option

Linux

Windows

MacOS

TANGO_USE_TELEMETRY

on

off

on

  • Some of the dependencies packaged with the binary wheels on PyPI have changed. The bundled versions are:

Dependency

Linux

Windows

MacOS

cpptango

10.0.4

10.0.4

10.0.4

omniorb / omniorb-libs

4.3.4

4.3.0

4.3.4

libzmq / zeromq

4.3.5

4.0.5-2

4.3.5

cppzmq

4.11.0

4.7.1

4.11.0

libjpeg-turbo

3.1.3

2.0.3

3.1.2

boost

1.87.0

1.86.0

1.87.0

abseil

20250814.1

-

20250512.1

protobuf

v33.5

-

v31.1

c-ares

v1.34.6

-

1.34.6

re2

2025-11-05

-

2025.11.05

OpenSSL

3.6.1

-

3.6.1

curl

curl-8_18_0

-

8.18.0

gRPC

v1.76.0

-

1.73.1

opentelemetry-cpp

v1.24.0

-

1.21.0

What’s new in PyTango 10.0.3?#

Date: 2025-07-25

Type: patch release

Here we provide short highlights of the changes. You can find the full changelog in revisions history.

Changed#

  • Silenced pydev debugger warnings about frozen modules at import time.

  • NumPy version pinned < 2.3 at build time. Newer versions can be used at runtime.

  • Updated logo for documentation.

  • OpenTelemetry SDK log level can be set via an environment variable.

Fixed#

  • Fixed regression in tango.DeviceProxy.read_attributes_reply() (noticed in Taurus).

  • Fixed segfault when client application using OpenTelemetry with OpenSSL exits quickly (under a few seconds).

  • Various issues in the Python DatabaseDS implementation.

Packaging#

  • PyTango packages are compiled with support for OpenTelemetry in Linux and macOS, but not Windows:

Compilation option

Linux

Windows

MacOS

TANGO_USE_TELEMETRY

on

off

on

  • Some of the dependencies packaged with the binary wheels on PyPI have changed. The bundled versions are:

Dependency

Linux

Windows

MacOS

cpptango

10.0.2

10.0.2

10.0.2

omniorb / omniorb-libs

4.3.3

4.3.0

4.3.3

libzmq / zeromq

4.3.5

4.0.5-2

4.3.5

cppzmq

4.11.0

4.7.1

4.10.0

libjpeg-turbo

3.1.1

2.0.3

3.1.0

boost

1.87.0

1.86.0

1.87.0

abseil

20250512.1

-

20250512.1

protobuf

v31.1

-

v31.1

c-ares

v1.34.5

-

1.34.5

re2

2025-07-22

-

2025.07.22

OpenSSL

3.5.1

-

3.5.1

curl

curl-8_15_0

-

8.14.1

gRPC

v1.73.1

-

1.73.1

opentelemetry-cpp

v1.22.0

-

1.21.0

What’s new in PyTango 10.0.2?#

Date: 2025-03-07

Type: minor release

Note: There was no 10.0.1 release (the number was skipped).

You can find the full changelog in revisions history with the links to related MRs, where the detailed descriptions of every change/new feature can be found.

Here we provide short highlights of the changes:

Changed#

  • Documentation updated with a new theme, and changed from reStructuredText syntax to MyST markdown.

  • Deprecated pipes for clients and servers. Removal scheduled for v10.1.0.

  • Deprecated the old experimental Tango object API which was broken. Removal scheduled for v11.0.0.

Added#

Fixed#

  • Occasional deadlock when a Group object that used events is destroyed.

  • tango.AttributeProxy.write_asynch() was broken.

  • Logging decorators like DebugIt work with show_kwargs=True.

  • Occasional segfault when anAttrConfEventData object are destroyed.

  • Segfault when pytest failure report tries to print device name.

  • Various issues in the Python DatabaseDS implementation

Packaging#

  • PyTango packages are compiled with support for OpenTelemetry in Linux and macOS, but not Windows:

Compilation option

Linux

Windows

MacOS

TANGO_USE_TELEMETRY

on

off

on

  • Some of the dependencies packaged with the binary wheels on PyPI have changed. The bundled versions are:

Dependency

Linux

Windows

MacOS

cpptango

10.0.2

10.0.2

10.0.2

omniorb / omniorb-libs

4.3.2

4.3.0

4.3.2

libzmq / zeromq

4.3.5

4.0.5-2

4.3.5

cppzmq

4.10.0

4.7.1

4.10.0

libjpeg-turbo

3.0.0

2.0.3

3.0.0

boost

1.87.0

1.86.0

1.87.0

abseil

20250127.0

-

20240722.0

protobuf

v29.3

-

v28.3

c-ares

v1.34.4

-

1.34.4

re2

2024-07-02

-

2024.07.02

OpenSSL

3.4.0

-

3.4.1

curl

curl-8_12_0

-

8.12.1

gRPC

v1.70.1

-

1.67.1

opentelemetry-cpp

v1.19.0

-

1.18.0

What’s new in PyTango 10.0.0?#

Date: 2024-10-01

Type: major release

An overview of the major new features are available in the Tango v10 and IDLv6 presentation materials from the Tango Collaboration meeting.

You can find the full changelog in revisions history with the links to related MRs, where the detailed descriptions of every change/new feature can be found.

Here we provide short highlights of the changes:

Changed#

  • PyTango requires at least cppTango 10.0.0. See the migration guide.

  • High-level Device and low-level LatestDeviceImpl classes now use Device_6Impl, with Tango IDLv6 interface

  • When using the Asyncio green mode, a DeprecationWarning will be emitted during the init_device() call, if any or the methods in Device are not coroutine functions (i.e., defined with async def). Note, support of synchronous methods in Asyncio servers will be removed in a future release. See the migration guide.

  • The Python GIL is released when adding/removing dynamic attributes calls into cppTango.

  • Error messages occurring during device start-up are now redirected to stderr instead of stdout

  • PyTango must be complied with the C++17 standard

Added#

  • Numpy 2.0 support

  • PyTango now supports OpenTelemetry for distributed tracing - see the telemetry how-to guide

  • Support for alarm events

  • Extended device information (DevInfo6 implementation, from IDLv6) - see the version info tutorial

  • For Asyncio green mode servers, added two coroutine functions to be used when adding and removing dynamic attributes. See the migration guide.

  • Pydevd debugging (as well as coverage) now is extended to dynamic attributes and commands. If necessary, the feature can be disabled by setting the environment variable PYTANGO_DISABLE_DEBUG_TRACE_PATCHING=1.

  • PyTango provides a stub file with typing information for improved autocompletion in your code editor (except conda packages for osx-arm64 and linux-aarch64)

  • Events can be pushed with Python exception objects directly, no need to convert to DevFailed

  • Device description, status, state can be set at the device start-up. See class attributes DEVICE_CLASS_DESCRIPTION, DEVICE_CLASS_INITIAL_STATUS, and DEVICE_CLASS_INITIAL_STATE.

Fixed#

  • Segfault when Restart Command is used on a PyTango server (first reported 6 years ago!)

  • Deadlock when pushing event from attribute read method (regression, introduced in 9.5.1)

  • Memory leak when writing DevString attribute with DeviceProxy is fixed

  • Segfault in push_archive_event(attr_name) with attr_name != state or status was fixed

  • Various fixes of asyncio tango.asyncio.DeviceProxy methods. High-level reads, and getattr calls can (and must) be awaited.

  • Fixes of *_asynch methods on DeviceProxy. I.e., when using Tango’s asynchronous push/pull callback model for accessing attributes and commands. This is not related to Python asyncio methods and coroutine functions.

  • class_property is now inherited by child device classes

  • Various issues in the Python DatabaseDS implementation

Removed#

  • Reverted the fix from 9.5.1 for resolving a crash in Asyncio devices when an attribute is read at the same time as an event is being pushed (original bug is not fixed). It was reverted because it caused a deadlock in existing code that was pushing events from attribute read methods.

  • Following cppTango, event type QUALITY_EVENT was removed from the EventType enum, as well as the quality_event_subscribed method from the Attribute class. See the migration guide.

Packaging#

  • PyTango packages are compiled with support for OpenTelemetry in Linux and macOS, but not Windows:

Compilation option

Linux

Windows

MacOS

TANGO_USE_TELEMETRY

on

off

on

  • Some of the dependencies packaged with the binary wheels on PyPI have changed. The bundled versions are:

Dependency

Linux

Windows

MacOS

cpptango

10.0.0

10.0.0

10.0.0

omniorb / omniorb-libs

4.3.2

4.3.0

4.3.2

libzmq / zeromq

4.3.5

4.0.5-2

4.3.5

cppzmq

4.10.0

4.7.1

4.10.0

libjpeg-turbo

3.0.0

2.0.3

3.0.0

boost

1.85.0

1.85.0

1.85.0

abseil

20240722.0

-

20240722.0

protobuf

v28.2

-

v27.5

c-ares

v1.33.1

-

1.33.1

re2

2024-07-02

-

2023.09.01

OpenSSL

3.3.2

-

3.3.2

curl

curl-8_10_1

-

8.10.1

gRPC

v1.66.2

-

1.65.5

opentelemetry-cpp

v1.16.1

-

1.16.1

What’s new in PyTango 9.5.1?#

Date: 2024-03-28 9.5.1

Type: minor release

Changed#

  • Restricted NumPy to 1.x, since we do not have NumPy 2.0 support yet.

  • Improved some error message related to invalid types passed to DeviceProxy.

  • Extended pydevd debugging and coverage to dynamic attributes and commands.

Fixed#

  • High-level attribute reads using asyncio DeviceProxies can now be awaited.

  • Asyncio green mode devices no longer crash when an attribute is read at the same time as an event is being pushed.

  • Numpy 1.20.0 no longer causes an import error.

  • High-level Device class inheritance now supports class_property.

What’s new in PyTango 9.5.0?#

Date: 2023-11-23

Type: major release

Changed#

  • PyTango requires at least cppTango 9.5.0. See the migration guide.

  • When using the asyncio green mode, a PyTangoUserWarning will be emitted during the init_device() call, if the user’s Device methods are not coroutines (i.e., defined with async def).

  • Use 127.0.0.1 as the default host for (Multi)DeviceTestContext instead of trying to find an external IP. This allows tests to work on systems that only have a loopback interface, and also reduces firewall warnings when running tests (at least on macOS). If using it from the command line like, python -m tango.test_context MyDS.MyDevice, an external IP is still the default.

  • All warnings generated by PyTango are now instances of PyTangoUserWarning, which inherits from Python’s UserWarning.

  • Some of the dependencies packaged with the binary wheels on PyPI have changed. The bundled versions are:

Dependency

Linux

Windows

MacOS

cpptango

9.5.0

9.5.0

9.5.0

omniorb / omniorb-libs

4.3.1

4.3.0

4.3.1

libzmq / zeromq

4.3.5

4.0.5-2

4.3.5

cppzmq

4.10.0

4.7.1

4.10.0

libjpeg-turbo

3.0.0

2.0.3

3.0.0

tango-idl

5.1.1

5.1.2

5.1.2

boost

1.82.0

1.83.0

1.82.0

Added#

  • Short-name access can be used for (Multi)DeviceTestContext devices. See migration guide

  • Experimental feature: use Python type hints to declare Device more easily. Read more in the new section: Use Python type hints when declaring a device.

  • set_write_value() now supports IMAGE attributes.

  • Forwarded attributes are partially supported in the (Multi)DeviceTestContext. We say partially, because a cppTango limitation (at least version 9.5.0) means root attributes on devices running in “nodb” mode (like those in launched by the TestContext) don’t work. However, it does work if the test device accesses a root attribute on a Tango device running with a Tango database.

  • Support for EncodedAttribute in high-level API devices.

  • Added free_it and clean_db arguments to tango.server.Device.remove_attribute() and tango.LatestDeviceImpl.remove_attribute() methods.

  • Support Tango server debugging with PyCharm, PyDev and VS Code. Breakpoints now work for command and attribute handler methods, as well as other standard Device methods, when running through a debugger that is based on pydevd. However, it doesn’t currently work with dynamic attributes and commands. If necessary, the feature can be disabled by setting the environment variable PYTANGO_DISABLE_DEBUG_TRACE_PATCHING=1.

  • Added support for Python 3.12.

Fixed#

Removed#

  • Breaking change to the API: the DevInt data type was removed, due to its removal from cppTango. See the migration guide.

  • Deprecated signature, get_write_value(self, lst)(), was removed.

What’s new in PyTango 9.4.2?#

Date: 2023-07-27

Type: minor release

Changed#

Added#

  • Correct code coverage of server’s methods can be obtained

  • server_init_hook was added to high-level and low-level API

  • macOS wheels now are provided

Fixed#

  • DevEncoded attributes and commands read methods are now segfault safe

  • DevEncoded attributes and commands now decoded with utf-8

  • DevEncoded attributes and commands can be extracted and written as str, bytes and bytesarray

  • If string encoding with Latin-1 fails, UnicodeError will be raised instead of segfaulting

  • When user gives empty spectrum properties to the DeviceTestContext, they will be patched with one space symbol “ “ for each element

  • In case patching failed or any other problems with FileDatabase, instead of crash PyTango will raise an exception and print out generated file

  • Regression when applying additional decorators on attribute accessor functions. Method calls would have the wrong signature and fail.

Removed#

What’s new in PyTango 9.4.1?#

Date: 2023-03-15

Type: major release (breaking changes compared to 9.4.0)

Changed#

  • Removed additional function signatures for high-level attribute read/write/is_allowed methods that were added in 9.4.0 resulting in a regression. For example, the high-level write method API for dynamic attributes of the form write_method(self, attr, value) has been removed, leaving only write_method(self, attr). Similarly, unbound functions that could be used without a reference to the device object, like read_function(), are no longer supported - only read_function(device). See the migration guide.

  • The dependencies packaged with the binary PyPI wheels are as follows:
    • Linux:

      • cpptango: 9.4.1

      • omniorb: 4.2.5 (changed since PyTango 9.4.0)

      • libzmq: v4.3.4

      • cppzmq: v4.7.1

      • libjpeg-turbo: 2.0.9

      • tango-idl: 5.1.1

      • boost: 1.80.0 (with patch for Python 3.11 support)

    • Windows:

      • cpptango: 9.4.1

      • omniorb: 4.2.5

      • libzmq: v4.0.5-2

      • cppzmq: v4.7.1

      • libjpeg-turbo: 2.0.3

      • tango-idl: 5.1.2

      • boost: 1.73.0

Fixed#

  • Regression for undecorated read attribute accessor functions in derived device classes. E.g., if we have class A(Device) with attribute reading via method A.read_my_attribute, then reading my_attribute from class B(A) would fail. More generally, repeated wrapping of methods related to attributes, commands and standard methods (like init_device) is now avoided.

  • Regression when applying additional decorators on attribute accessor functions. Method calls would have the wrong signature and fail.

What’s new in PyTango 9.4.0?#

Date: 2023-02-15

Type: major release

Warning

significant regressions - use newer release!

Changed#

  • PyTango requires at least cppTango 9.4.1. See the migration guide.

  • Breaking change to the API when using empty spectrum and image attributes. Clients reading an empty attribute will get an empty sequence (list/tuple/numpy array) instead of a None value. Similarly, devices that have an empty sequence written will receive that in the write method instead of a None value. See the migration guide on empty attributes and extract as.

  • Python dependencies: numpy is no longer optional - it is required. Other new requirements are packaging and psutil.

  • Binary wheels for more platforms, including Linux, are available on PyPI. Fast installation without compiling and figuring out all the dependencies!

  • The dependencies packaged with the binary PyPI wheels are as follows:
    • Linux:

      • cpptango: 9.4.1

      • omniorb: 4.2.4

      • libzmq: v4.3.4

      • cppzmq: v4.7.1

      • libjpeg-turbo: 2.0.9

      • tango-idl: 5.1.1

      • boost: 1.80.0 (with patch for Python 3.11 support)

    • Windows:

      • cpptango: 9.4.1

      • omniorb: 4.2.5

      • libzmq: v4.0.5-2

      • cppzmq: v4.7.1

      • libjpeg-turbo: 2.0.3

      • tango-idl: 5.1.2

      • boost: 1.73.0

  • When using the --port commandline option without --host, the ORBendpoint for gio::tcp passed to cppTango will use "0.0.0.0" as the host instead of an empty string. This is to workaround a regression with cppTango 9.4.1. Note that if the --ORBendPoint commandline option is specified directly, it will not be modified. This will lead to a crash if an empty host is used, e.g., --ORBendPoint giop:tcp::1234.

Added#

  • User methods for attribute access (read/write/is allowed), and for commands (execute/is allowed) can be plain functions. They don’t need to be methods on the device class anymore. There was some inconsistency with this previously, but now it is the same for static and dynamic attributes, and for commands. Static and dynamic commands can also take an fisallowed keyword argument. See the migration guide.

  • Device methods for reading and writing dynamic attributes can use the high-level API instead of getting and setting values inside Attr objects. See the migration guide.

  • High-level API support for accessing and creating DevEnum spectrum and image attributes. See the migration guide.

  • Developers can optionally allow Python attributes to be added to a DeviceProxy instance by calling unfreeze_dynamic_interface(). The default behaviour is still to raise an exception when accessing unknown attributes. See the migration guide.

  • Attribute decorators have additional methods: getter(), read() and is_allowed(). See the migration guide.

  • Python 3.11 support.

  • MacOS support. This is easiest installing from Conda-forge. Compiling locally is not recommended. See the installation guide.

  • Integrated development environment (IDE) autocompletion for methods inherited from tango.server.Device and tango.LatestDeviceImpl. Attributes from the full class hierarchy are now more easily accessible directly in your editor.

Fixed#

  • Log stream calls that include literal % symbols but no args now work properly without raising an exception. E.g., self.debug_stream("I want to log a %s symbol"). See the migration guide.

  • Writing a numpy.array to a spectrum attribute of type str no longer crashes.

  • Reading an enum attribute with ATTR_INVALID quality via the high-level API now returns None instead of crashing. This behaviour is consistent with the other data types.

Removed#

  • Support for Python 2.7 and Python 3.5.

  • The option to install PyTango without numpy.