I was right about the double slash bug [1]—it was a simple bug after all [2]. The authors of two Gemini crawlers wrote in about the double slash bug, and from them, I was able to get the root cause of the problem—my blog on Gemini [3]. Good thing I hedged my statement about not being the cause yesterday. Sigh.
Back in Debtember, I added support for displaying multiple posts [4]. It's not an easy feature to describe, but basically, it allows one to (by hacking the URL (Uniform Resource Locator), but who hacks URLs these days?) specify posts via a range of dates. And it's on these pages that the double slashed URLs appear. Why that happens is easy—I was generating the links directly from strings:
local function geminilink(entry)
return string.format("gemini://%s%s/%s%04d/%02d/%02d.%d",
config.url.host,
port, -- generated elsewhere
config.url.path,
entry.when.year,
entry.when.month,
entry.when.day,
entry.when.part
)
end
instead of from a URL type. I think when I wrote the above code, I wasn't thinking in terms of a URL type, but of constructing a URL from data I already had. The bug itself is due to config.url.path ending in a slash, so the third slash in the string literal wasn't needed. The correct way isn't that hard:
local function geminilink(entry)
return uurl.toa(uurl.merge(config.url,
{
path = string.format("%04d/%02d/%02d.%d",
entry.when.year,
entry.when.month,
entry.when.day,
entry.when.part)
}))
end
and it wouldn't have exhibited the issue.
With this fix in place, I think I will continue to reject requests with the double slash, as it is catching bugs, which is a Good Thing™.