Provides a simple CSP appender where URIs are appended to allowed keys. At the moment, this is most useful for allowing custom URIs.

Necessary to encapsulate a single browser instance which can remain in memory without creating a new instance per screenshot. This will reduce the number of instances created. Here’s what was observed on macOS prior to this change:

• Google Chrome: ~128MB per instance.

• Google Chrome Helper: ~26MB per instance (with ~three per Google Chrome instance).

After this change you’d get an instance which hovers around ~100MB with supporting helpers that would consume the same amount of memory but clear up after rendering process completed.

This also introduces a few new features:

• When calling, a block must be used which automatically closes the page when finished.

• The instance will be restarted after a maximum number of renders.

• An empty success is returned only when no failures are encountered.

Will soon be used by Screens::Shoter to render screenshots.

Reek was updated to ignore the multiple statements in the #call method since it’s worthwhile to keep the success and failure logic in a single method.

Necessary to validate a list of URIs which must all be of https scheme. This will soon be used for building a list of allowed domains for Content Security Policy purposes.

Reek was updated to ignore the multiple constants for types as we want to keep all of these within the same module.

Necessary for folks to supply custom URIs (domains) which will be appended to the application Content Security Policy.

Necessary to provide UI support for bulk importing of devices from a CSV file. This includes new translations, route, action, templates, and linking via the bulk upload SVG icon at the top of the devices index template.

Necessary to provide a distinct icon for bulk imports.

Necessary to explain what this is and why it’s important.

None.

Necessary to provide a template folks can download for local use and customization. This shows how to import a Demo device for the TRMNL OG (2-bit) model.

Ensures data being imported for device provisioning is valid.

Provides single device validation.

Necessary for bulk import of multiple devices at once by using a CSV of multiples devices as the source. See [Ruby CSV Parsing](https://alchemists.io/articles/ruby_csv_parsing) for futher details.

Necessary to validate allowing content to be imported from multiple sources:

• Attachment: Allows import via a CSV file attachment.

• URI: Allows import via a remote URI.

Only one or the other is allowed, not both at the same time.

A quality of life improvement by making it quick to flush the keystore database, especially for development purposes.

Provides convenience for finding an existing record by name or creating a new record if not existing.

Will be used to resolve concurrent read/write locks, especially with the Ferrum Browser.

Provides dynamic configuration of the Content Security Policy based on allowed URI settings as computed by the CSP Appender. Unfortunately, Hanami does not provide direct access to the policy for updating so I’m reaching into the private policy instance variable to manipulate directly. Definitely not ideal but keeps the code tight.

Both i18n and client objects are extremely verbose when inspecting objects so this cleans that up so the console is easier to read.

Allows domains to be ignored, when empty, for situations where you might want to allow any URI.

Reek was updated to ignore the multiple statements for the check because it’s important to keep all checks together.

Details can be found here:

• htmx Remove 4.2.0.

• htmx Slide 0.8.1.

Weekly maintenance and care.

Minor clarification to indicate this script is for production while bin is for development.

Reduces duplication by delegating to the new browser object. This also cleans up RuboCop and Reek accordingly.

Details.

None.

Details. This also picks up a patch to Down.

Details.

Details. This introduces a breaking change because the error messages now have the wrong encoding. To account for this the effected specs were updated accordingly because my String refinement only alters the first character whereas String#capitalize doesn’t. Again, I this is a bug with the JSON gem.

Details.

Details. This also picks up a few gem patches (along with a minor update to Rack Test).

Details. This also picks up Rodauth 2.49.0.

Details.

Necessary to match what is documented.

Necessary to clarify you need your API Key when authenticating.

Necessary to speed up conflict resolution because UnusedParameters tends to duplicate what is reported by RuboCop so you get hit with the same issues by both linters when only having RuboCop is enough for reporting and resolution purposes.

Ensures clone and download links are scoped to their corresponding extension cards so multiple actions avoid click ambiguity. This was originally reported by Simon Fricker.

Accidentally forgot, and noticed after looking at the previous night’s logs, that the renders count wasn’t being reset back to zero. This accounts for that.

This also increases the maximum to 500 to give the browser a bit more time (although what the "best" limit is yet to be determined).

This introduces a read/write lock to allow multiple concurrent readers but only one concurrent writer in order to ensure we can safely restart the browser once maximum renders or recoveries are reached. This forces the multiple readers to wait until the write lock is released.

Related to the above, this also applies the following changes:

• Adds a small maximum recovery limit for when a dead browser is detected so we can recover by restarting the browser.

• Ensure we not only log a fatal message with a ProcessTimeoutError but re-raise the error in order for the server to be restarted since we can’t do anything with a client that is nil.

• Updated the dead browser log to show a warning so you are alerted that recovery took place but you might also want to keep tabs on further issues.

• Updated all debug messages as error messages so when these errors occur they are more apparent and to help indicate more action might be required.

RuboCop was updated to ignore method length for #call since there’s not an elegant way to break up the rescue blocks further.

Switched to using a regular expression because, sometimes, the users count can toggle between zero and one…​but we only care that the link exists with a number and a URI which this now expresses.

Improves consistency since these were the only outcasts that didn’t list these form fields at the top like all other resources in the system.

Leans on object equality since we are using whole value objects. This allows us to use exact matches instead of fuzzy matches.

Ensures, when playlist ID is present, the given playlist is used instead of creating a new one.

Ensures block content — when block is missing — doesn’t override the message with nil content by allowing block_content to fallback to the message when a block isn’t given. This also removes the combined local variable since we can splat all attributes at once instead of creating an intermediate local variable first while reducing object allocation further.

A feature spec was added to account for this which, indirectly, tests the logger patch.

We already have the page content in memory, so we can skip writing to a temporary file, entirely, by passing the content straight to the browser directly. This also ensures multiple temporary directories aren’t left around which avoids further cleanup and management by this object.

Added the hash refinement so you can create an instance from the console to avoid getting an error that the refinement couldn’t be found.

Reek was updated since the data clump issue is fixed but added an excluding for duplicate call to instance#quit since Reek doesn’t distinguish the importance of the ensure block.

Necessary to resolve deprecation warnings since use of the Hash#use refinement will be removed in the next version.

This is small enough to fold into the main configuration now that most internationalization issues have been addressed.

This was taking up unnecessary screen real-estate so have removed in favor of having more focus on the image being presented. Plus this reduces another asset from being loaded since the SVG isn’t used elsewhere.

Due to the above change, this also adds, a subtle, dark grey border to the image which provides a nice visual aid for images that are smaller (i.e. wide screen) to show the space they don’t occupy.

Due to a device always being associated with a model, there is no reason to store separate width and height attributes since they can be derived from the model. The following changes make this possible:

• The schemas were updated to ignore width and height attributes.

• The device part was updated to delegate to the associated model for dimensions.

• The device serializer was updated to ensure model dimensions are included.

• All Devices API actions (and the provisioner) were updated to ensure a model is associated with a device.

• The Devices API delete action now answers a not found problem detail when a device can’t be found since the serializer always needs a device associated with a model.

• The database migration removes the width and height columns from the device table.

Safe to remove since this is implicit behavior and being explicit adds no value.

Safe to remove since the superclass provides this.

Redundant and safe to remove since the version is displayed at the top.

We only need to return in both of these cases.

Allows these rules to be more dynamic — and reusable — within different structures where the root key might differ. This still assumes the rule is one level deep, though.

Clarifies this translation is for browser restarts. Fixed the keys so they are alphabetically sorted for quick lookup.

Reduces duplication since we can delegate to the repository for finding or creating records.