Lines taking from the asynchronous processing of stdout/stderr where
left as str, leading to encoding problems when utf8 characters showed
up in the stream and were being interpreted as ascii encoding.
(cherry picked from commit 9373be3)
When using the version cache only use the version cache if the OctoPrint
version stored within it matches the one of the currently running
instance. Otherwise we might report false positives with regards to
available updates under some circumstances.
(cherry picked from commit bb7b0cb)
Should be valid zip archives/tarballs. Also use only the extension
of an uploaded plugin archive as suffix for the temporary file that's
used for installing it.
(cherry picked from commit a1ff698)
Plugin Manager now internally uses pkg_resources.parse_version and
pkg_resources.parse_requirements to check for version compatibility
with plugins from the repository. That will allow to directly use
the OctoPrint version string once it's PEP440 compatible.
(cherry picked from commit 8eb61a9)
Both file search and plugin repository search reloaded the page when
hitting enter in the search field. This fixes that.
(cherry picked from commit 60043cc)
Local file storage implementation will now "touch" added files to set
modification date to current date, ensuring that files that are just
moved through a custom implementation of file object (e.g. through the
watched folder mechanism) will show with "now" as the upload date.
(cherry picked from commit 5f0c89c)
If compatibility information is provided only as a version number it's
prefixed with >= for the check (so stating a compatibility of only
1.2.0 will now make the plugin compatible to all versions that follow
too).
Alternatively the compatibility information may now contain stuff like
">=1.2,<1.3" in which case the plugin will only be shown as compatible
to OctoPrint versions 1.2 and up but not 1.3 or anything above that.
(cherry picked from commit d5692d7)
If the SD card fails to initialize with a volume.init failure, that's
not a reason to break off all diplomatic relations with the printer.
(cherry picked from commit 9caf13d)
Custom exception should be derived from Exception, not BaseException.
Not only is this specified in the python documentation, but also
tornado will be able to handle Exceptions in requests perfectly fine
and return an HTTP 500 for them, but crash hard (as in, server shut
down follows) for BaseExceptions.
(cherry picked from commit 29b047b)
That way plugins don't need to implement the SettingsPlugin mixin in order to access their data folder, which is now an injected property in OctoPrintPlugin. Plugin authors should still always use the getter since that will also make sure the folder actually does exist (lazy creation).
PluginSettings.get_plugin_data_folder has been marked as deprecated.
Doesn't work reliably across pip versions, and adding a special rule for each and every version also isn't fun. Plugin authors will need to use a different approach for pulling in dependencies that are not on pypi during install, e.g. extra install urls or something like that. For now, I give up here ;)
Will lookup the correct pip command to use if not provided as constructor parameter and determine the version. Implements __lt__, __le__, __gt__ and __ge__ for "less/greater than" checks against the pip version, for dynamicalle changing call parameters provided to the execute method based on availability.
Will allow adjusting software update and plugin manager plugins to only use --process-dependency-links for pip versions >= 1.5
Reduces number of requests done against (rate limited) github API, especially after consecutive server restarts. Forcing the update check still will query the API.
That's also able to cope with 0.3a and similar "non-semantic" version schemes. Semantic version comparison can still be forced by setting the compare_type in the check to "semantic". Strict equality check can be set similarly by using the value "unequal".
That's more user friendly than just blindly reloading the page. Also include installed Template- and AssetPlugins in the check whether there's a new version or not.
Caused by a wrong variable usage. Fixed the usage and added a migration
step that fixes the circular reference if it exists. Bumping configuration
version to 2.
Fixes#946
Introduced new callback for settings plugin to hook into the point after the settings have been migrated since the initialize method is not suitable for that as it might be called before other plugin implementations reachable through hooks are initialized.
Usage of super(MyClass, self) is henceforth discouraged due to dynamic reloading within OctoPrint's plugin subsystem. Refer to https://thingspython.wordpress.com/2010/09/27/another-super-wrinkle-raising-typeerror/ for details on this problem.
Action by plugin authors is needed to remove the super(...) call and substitute it with octoprint.plugin.SettingsPlugin.on_settings_save(self, data) and octoprint.plugin.SettingsPlugin.on_settings_load(self). Without these modifications, calls to plugin methods utilizing super(...) will fail after a call to the plugin managers reload_plugins method.
Closes#942
The SoftwareUpdate Plugin had a bug that caused way too much check data to be stored in the configuration, leading to plugins potentially being stuck in an "update available" loop although the update had already been applied. Now only the current version of github_commit update types is persisted, not the full check configuration.
Also introduced a configuration version and made the migration function migrate old configs to remove anything that was same as the default supplied for the "octoprint" and all plugin hook checks. That should clean things up in existing installations.
OctoPrint will take care of calling a migration function on the plugin if the plugin demands a newer configuration version than currently stored in config.yaml.
When any plugins are enabled/disabled, the client now gets a trigger to fetch a fresh list of update information from the server. This should take care of any old popups still hinting at now uninstalled plugins.
This is to make sure that they all end on a ; and hence stuff doesn't break if a file included directly after starts with a (...) construct. Of course it would be better if all bundled files are valid in that matter, however since we can't enforce that, we'll add this slight overhead to reduce the risk of stuff breaking.
See for example: http://stackoverflow.com/questions/20307462/js-cant-combine-lib-files
python-setuptools on OctoPi is quite old and trying to update in a local virtualenv apparently CAN cause problems (we had this with python-serial too a while back), so a backwards compatible solution is better here.
If plugins get enabled or disabled, the update check configuration needs to be refetched next time it's needed since there might have been changes to plugin implementing the check_info hook.
Same holds true for cache files, so there should be no problem anymore with installs where the static folder is not writable.
Also introduced two new devel config vars to disable merging and minifying of the assets, solved the empty-less-bundle problem and made sure babel knows about the jinja extension.
This should allow users to adjust the behaviour of the communication stack more granularly in case the regular settings which only consider G4, G28, G29, G30, G32 to be real long runners without output are not sufficient.
Actual test whether key is provided or not happens in before_request handler of flask app, hence the more granular processing in the decorator was unecessary since around november '14. Adjusted to only do the actual processing necessary to replicate login_required behaviour/login users if necessary.
Closes#516 (actually, that was solved since introduction of the before_request handler which took that work off the hands of the decorator, but now it's properly documented too)
Match against entry_point origins using both package name and package name plus version. If one of those matches (depends on pip version if output will include version or not), it's the plugin we are looking for.
This way we don't need to try to strip a version number from the package name as output by pip without knowing if a version number is included or not.
Some pip invocations return extra info after the Successfully installed
(like "Cleaning up..."). Also, not all packages include a version
number in their package names.
The only translation that is going to be bundled with OctoPrint for now will be .de, if at all. The reason simply being that this is the only translation I can update myself and hence guarantee a good user experience for.
setup.py now offers two new commands, babel_bundle and babel_pack, with which translations for plugins and core OctoPrint can be automatically bundled or packed as language packs once compiled. This should make sharing translations quite easy in the future.
That is the first step towards downloadable (and separately managed) translation packages, decoupled from the software's release cycle and source control. Since it will be impossible to properly maintain all languages concurrently with OctoPrint's releases, decoupling that will hopefully increase the user experience through preventing broken translations (or stalling releases due to having to wait for the translations to be updated).
Sometimes they will still get discovered by python even though their packages have since been uninstalled.
This will also lead to them being reloaded after an uninstall and a subsequent plugin reload.
Marking them as uninstalled and not handing out uninstalled plugins when collecting them solves this.
Added phase specific hooks for queuing, queued, sending and sent phases of a command ("octoprint.comm.protocol.gcode.<phase>"). Removed old queuing phase hook and declared as obsolete hook in plugin manager to prevent plugins that depend on it from being enabled.
Adding those new hooks also necessitated refactoring the whole command processing, made it more modular and added phase specific handler functions that allow handling all blocking commands centrally for example.
Since it cannot be reliably determined by the system if a hook is essential for a plugin's functionality or not, it makes more sense to just disable plugins that utilize obsolete hooks then risk running half working plugins.
This will hopefully prevent conflicting routes between multiple plugins from being registered and also ensures a more consistent behaviour compared to BlueprintPlugin mixins.
Thanks to @Salandora for bringing this up.
Hidden files (starting with .) will produce a 404. This also excludes .metadata.yaml from being downloadable. The alternative of adding all extensions defined by plugins to the regex dynamically was not chosen since that would necessitate to make plugins implementing the "octoprint.filemanager.extension_tree" hook restart needing plugins in the lifecycle management for a minimal gain in perceived security.
Solves #897
Added a setting to allow -90 degree rotation. Combined with fliph and
flipv, I think this means every one of the 8 major orientations is covered
(2^3). 0, 90, 180, 270 times 2 (mirror image in each rotation). I chose
-90 instead of 90 because that keeps the upper left corner pinned to the
same spot.
The extra ms-transforms aren't for this issue exactly but allowed me to
test my changes in ie. I've tested Chrome, Firefox and IE11. I don't
have Safari handy, but if it breaks it *should* be only when this is
enabled.
Otherwise we get import issues during build since the octoprint module depends on the dependencies already being present what they obviously can't be at the beginning of setup.
Babel related commands and the Clean command for setup.py are not reusable by plugins as well. Also added a factory method for plugin setup parameters.
The new hook allows extending the list of rules for maximum body sizes differing from the default of 100KB and can be used by plugins to allow uploads to them that exceed that file size.
Also extended the plugin manager to detect plugins that implement restart needing hooks (such as the above one) and handling those plugins the same as plugins containing implementations that inherit from octoprint.plugin.core.RestartNeedingPlugin
Have you ever noticed when developing that every time you stop and start
the server, the terminal window gets an extra duplicate line for every
reconnect attempt? Well, it's because (I think) "delete" in javascript
just removes the indicated name from the namespace, it doesn't actually
free up an object. Those zombie objects are still there and wake up (for
some transports) on reconnect. Might be different in SockJS v1 or later.
Cache clearing doesn't work reliably with location.reload(true), leading to cache induced infinite loop when disabling access control (since generated template still assumes first run dialog to be active). Disabling for now, need to find a better way to reload the page while clearing the browser's and server's cache (same issue exists for plugins needing a client side triggered reload).
Fixes#886
The server now sends the current server time with each temperature update as well as the initial temperature data, allowing the client to translate the timestamps contained in the temperature logs to its local timescale.
Followup to commit f19d0f0360 which caused a problem with large deviations between server and client time.
Cut off of the temperature graph is now not based on the number of data points any more but on the actual time of the data points. Anything older than ``n`` minutes will be cut off, with ``n`` defaulting to 30min. This value can be changed under "Temperatures" in the Settings
Closes#343
Sending "!!DEBUG:action_custom foo 1 2 3" will result in the virtual printer sending "// action:foo 1 2 3" back. Sending "!!DEBUG:action_custom bar" will result in the virtual printer sending "// action:bar" back.
If only one tool is configured, the profile dialog now won't show the offset configuration. If more than one extruder is configured offset configuration for anything but the first tool (which acts as reference for the relative offsets of the others) will be shown.
Closes#677
Instead of matching to full strings send by firmware in case of checksum or line number issues, now matches only against lower case "checksum", "line number" and "line expected". Should make it more resilient to firmware side changes of the sent messages.
jQuery bubbles all children show/hide events up to the parent so the
event gets called whenever a settings pane is chosen by the user.
Intent of the viewmodel callback was just on first show and final
dismiss.
Any unknown commands should just be piped through and not use up an acknowledgement. This
new behaviour can be overridden via the new feature flag "unknownCommandsNeedAck", which will
restore the former behaviour causing even unknown commands to use up an "ok".
Also no line numbering or checksumming will happen. Shouldn't usually be of relevance for other
firmwares, but in case of any compatibility issues introduced by this also added a new feature
flag "sendChecksumWithNonGcode" that causes even unknown commands to be sent with a
checksum if necessary.
Repetier always first sends the ok and then any command output. In case of M105, that makes the response look like from an externally triggered heatup (no ok on the same line), causing polling to stop until that falsely detected heatup is complete. Added a configuration option to disable heatup detection. The disadvantage of this is that when printing via Repetier Firmware from SD, the heatup times won't be substractable from the total print time, leading to a less accurate print time left estimation.
Closes#835
Removed old loginstate.subscribe method and migrated existing code to new mechanism.
This should allow plugins to only request their sensitive data if a user is a) logged in and has b) the necessary rights for a request. The server should still do additional auth checks, but this way no unnecessary request have to be made.
* Renamed userjson to userdata, removed original userdata (developers will live if they need to ensure their supplied userdata is valid JSON)
* Input validation for userdata parameter - a HTTP 400 response will be generated if userdata can't be parsed as JSON
* Adjusted documentation accordingly
* moved virtual printer into plugin
* made default serial factory use supplied parameters instead of directly utilizing self._port and similar
* documented new hook
f98ebaaf moved the passive_logon code to a new function from
api/__init__.py to util/flask.py. Before that move any fallthrough
would hit return NO_CONTENT, but now it returns nothing which makes
flask complain.
This hook allows the plugin to create the serial port object (or pass to
the next hook or the builtin serial port handler) which allows for
serial communication filtering or a different communication transport.
Use @foosel's code to use the same loop for hooks as well as the
builtin.
Also, modify the cura slicer to recognize the setting and avoid
adjusting the model position, just leave it in the center where it would
be on a circular platform
Having to wait for the first serial timeout before sending the first M105 even when not waiting for seeing a "start" caused unnecessary wait times for reaching the "Operational" state.
Add computed property "name" to function objects if not provided by the
browser, to avoid dropping out of main.js before it is finished.
(cherry picked from commit 68215c6)
Tracking the "ok" attached to a resend is necessary in order to make sure that it does not trigger the resend of the next line after the requested one right after, causing resend loops in some cases, e.g. "> 100", "> 101" , "< rs 100", "ok", "> 100", "< expected 100, got 101, rs 100" -- here the last error from the firmware could not be processed as "false negative" correctly (101 was already sent to the printer when it detected the error for 100, so this error just needs to be ignored) since the resend flag was already cleared due to line 100 and 101 having been enqueued, 100 in the resend handler, 101 due to the following ok. This patch fixes the latter, thus solving the problem.
Plugins may be enabled and disabled during runtime. If they are of types which allow hot loading, this will be done. Otherwise they will be marked as pending and updated after a restart. Same for installation and uninstallation.
Plugins may be loaded, unloaded, activated and deactivated. Errors while trying to load a plugin or initializing an implementation will only result in it staying deactive but registered in the system, allowing it to be further processed e.g. by a plugin manager
Can be used to always retrieve the defaults (by providing an empty config dict) or to utilize the get method with other settings than the system settings.
Multiple mixins are allowed of course. Allowing multiple implementations lead to too many problems due to plugin names for referring to the APIs of SimpleApiPlugins or the assets of AssetPlugins.
Hence __plugin_implementations__ has been deprecated in favor of __plugin_implementation__. The plugin subsystem will automatically copy the first implementation from __plugin_implementations__ to __plugin_implementation__ and log a deprecation warning.
Adjusted documentation accordingly. Also added docs for helpers.
Detected plugins are now logged in a better readable way and with additional information.
Also marked folder for bundled plugins as bundled - forgot to do that earlier.
This way user adjustments to those timeouts via the settings
will take immediate effect even while the connection to the
printer is already established.
