[–] 2 points 7 months ago*

So, assuming good faith, they used two Telegram bots for some service functionality

these two bots are used to resolve username from user id, eg tg://user?id=25

Obviously, that should never happen silently. But these findings don't necessarily mean data has been compromised [beyond the scope of the app itself].

I get they may be very frustrated and annoyed at the negative blowback after their FOSS efforts, but dismissing concerns isn't a good way to respond.

  • source
  • [–] 3 points 7 months ago*

    How does Pretext work?

    1. Segment the text; Normalize whitespace, apply Unicode line-break rules, and split the string into measurable units using the browser's own text segmentation.
    2. Measure with Canvas; Feed each segment through Canvas measureText() to get real glyph advance widths from the font engine. Results are cached.
    3. Pretext.js uses pure arithmetic; Given a container width, compute line breaks by summing segment widths. Multiply line count by line-height. Return height. No DOM, ever.

    Unfortunately, that doesn't really explain the final integration. And it seems I misunderstood/-assumed at first.

    Looking at the example at the top right, it renders numerous div elements?

    So, presumably, you lose text wrap behavior and clean markup like <p> for a paragraph? I also can't select text from it in a normal or consistent way.

    This example isn't very convincing either.

    Seems like a cool visual gimmick more than practically useful and accessible for primary content.

  • source
  • [–] 3 points 7 months ago

    I work for a small ~30-person company with various customers, including some very big names. We're very deliberate about where tools like those could help us, where it's worth the exploration and investment. We want to be innovative and have the expertise, but at the same time, be reasonable and sound. We're also very conscious of data sharing and safeguards, in part out of necessity, because we can't just share our customers' code or data with third parties.

    Excitement, commitment, use, and hopes of using AI tools differ between colleagues. What we can use and how differs between projects.

    So yes, there are definitely other kinds of companies and environments out there.

  • source
  • [–] 1 point 7 months ago

    Retrospectives are great for finding and sharing a consensus on these kinds of issues. The team can weigh their options. Known limitations are much better than unknown ones. And often, some bandaids and workarounds are possible to diminish negative effects, at least to a degree.

    I've definitely had things we had to wait for, or are still waiting for. At least we don't usually get outright rejections.

  • source
  • parent
  • context
  • [–] 46 points 7 months ago (10 children)

    A code leak doesn't give a code and product use license. Any project and product use based on the leaked code is less stable and safe than other solid projects under clear terms. OpenCode is not obsolete.

  • source
  • view more: ‹ prev next ›