weixin_39856630
weixin_39856630
2020-11-22 16:57

Bump treetop to 1.6

Treetop 1.5 is now five years old and some distributions have been running with treetop 1.6 without issue for some time.

该提问来源于开源项目:asciidoctor/asciidoctor-pdf

  • 点赞
  • 写回答
  • 关注问题
  • 收藏
  • 复制链接分享
  • 邀请回答

6条回答

  • weixin_39856630 weixin_39856630 5月前

    would you mind considering this PR please? I have no data to back up the performance hit but wondering if this can be re-visited.

    点赞 评论 复制链接分享
  • weixin_39890327 weixin_39890327 5月前

    We've held back the version of treetop because treetop 1.6 is slower. I don't know why it's slower, but it's substantially slower. So if we can answer why it is slower, and whether it will ever be faster than 1.5 was, then I'm ready to proceed. That's the hump we need to get over.

    点赞 评论 复制链接分享
  • weixin_39890327 weixin_39890327 5月前

    I've tested it again and it's still slower. Here's the difference in a test parse of 50 runs.

    
    1.5.3 - 0.001674898
    1.6.10 - 0.002010059
    

    That's an increase of 20%, which is huge. Given that speed is already an issue when creating PDFs, I'm very hesitant to add overhead for no good reason.

    点赞 评论 复制链接分享
  • weixin_39890327 weixin_39890327 5月前

    It does appear that the parsing is such a small part of what Asciidoctor PDF does that it doesn't impact the speed of the test suite. So while this increase is huge, it's impact doesn't seem to be noticeable.

    What I don't understand is why treetop got so much slower. It still generates the same parse code. So where's the overhead coming from? Are they doing anything to address it?

    点赞 评论 复制链接分享
  • weixin_39986027 weixin_39986027 5月前

    it's impact doesn't seem to be noticeable

    So, is there something that prevents this from being merged then?

    点赞 评论 复制链接分享
  • weixin_39890327 weixin_39890327 5月前

    Frankly, I just don't care anymore about the line I was trying to hold on this, so let's just upgrade.

    点赞 评论 复制链接分享

相关推荐