WeblateOrg/weblate

Describe the problem The Anthropic machinery sends the system prompt as a plain string, without any cache_control block. Anthropic's prompt caching is opt-in, so nothing gets cached at all right now. The system prompt built in BaseLLMTranslation (persona + style + language instructions + the JSON schema boilerplate) is identical for every request to a given target language. With a longer style guide it easily reaches 2-3k tokens. When automatic translation runs over a few thousand strings, that same prefix is billed at full input price on every single request. In our setup (~3.4k strings per language, ~8KB style prompt) this repeated prefix is the majority of the MT spend. Solution brainstorm Send the system prompt as a content block with a cache breakpoint in weblate/machinery/anthropic.py: "system": [ {"type": "text", "text": prompt, "cache_control": {"type": "ephemeral"}}, ], Cached reads are billed at roughly 10% of the normal input price with a 5 minute sliding TTL, which fits

Describe the problem In Godot Engine's class reference, we make great use of BBCode to format text. A lot of go beyond the standard [b], [i], [code] tags and require additional parameters. For example, the [url] tag can optionally accept a link, formatted as [url=https://github.com/WeblateOrg]. Our custom [codeblock] tag can also accept a language, formatted as [codeblock lang=csharp]. Unfortunately, the "highlighting" and "error reporting" systems cannot detect these tags. This can make it quite awkward to change several strings. BBCode-related changes are harder to spot, tags may be mistyped or missing in translations, shortcuts such as CTRL + number key to quickly insert tags do not work. Solution brainstorm I believe what mostly needs to change is this regex in this repository: weblate/weblate/checks/markup.py Lines 76 to 78 in 2b4eab3 BBCODE_MATCH = re.compile(

Describe the problem The keyword option in gettext can be defined multiple times to define more keywords. The addon currently only allows defining one additional keyword. Solution brainstorm Allow defining a list of keywords and add the --keyword option for each of these keywords. Describe alternatives you have considered No response Screenshots No response Additional context weblate/weblate/addons/gettext.py Line 1123 in 03f08b9 result.append(f"--keyword={keyword}") I can try implementing this if desired.

Describe the problem As a maintainer of translations, I sometimes want to ask contributors about their contribution. That can be done through the Comments tab. At the moment when I start typing a @ symbol for the mention, no suggestions appear. I Suggestions appear when I start typing, but that seems to search through the whole list of known Weblate users. This means that when I'm on the History tab and see a user I want to discuss with, I have to take note of the entire name, because I can't view it once I moved over to the Comments tab to start to create a comment with a mention. Solution brainstorm I'd like to have suggestions based on my project, the language and the string. There are about 50 people that contributed to the whole translation project, 5 that contributed to that language and maybe less that worked on that string. It would make sense to highlight those users first. This is also the behavior I've become accustomed to from other collaborative software like git hosting a

Describe the problem DELETE /api/projects/{slug}/ returns 204 No Content immediately, but the actual project removal runs asynchronously in a background task. A client has no way to distinguish "fully deleted" from "deletion merely queued" from the response alone — the only way to find out is to poll GET /projects/{slug}/ until it 404s, or dig through the change/audit log after the fact. This is workable but easy to get wrong: automation that treats 204 as "safe to proceed" (e.g. before recreating a resource, or reporting the operation as complete to a user) can act on stale assumptions about project state while the real deletion is still in flight. Solution brainstorm Return 202 Accepted instead of a bare 204, with a task/status URL in the response body — Weblate already has exactly this machinery for other async operations via /api/tasks/{id}/. A client can then poll the task URL to know precisely when removal has actually completed, rather than inferring it indirectly. Describe