QUERY_STRING should not contain a leading question mark (?)

[dernst]

As per the title httpuv produces QUERY_STRING values containing the leading question mark of the form ?foo=bar where it should be foo=bar.

According to the documentation httpuv follows the Rook specification which states that "QUERY_STRING: The portion of the request URL that follows the ?, if any. [...]" [1]

Rook itself is modeled after CGI which also states that the question mark is not part of the query string and provides the following BNF syntax [2]:

   script-URI = <scheme> "://" <server-name> ":" <server-port>
                   <script-path> <extra-path> "?" <query-string>

The current behavior of httpuv is surprising and not consistent with established standards.

[1] https://github.com/jeffreyhorner/Rook
[2] https://tools.ietf.org/html/rfc3875#section-3.3

[wch]

I think you are correct that the QUERY_STRING should not contain the leading ?.

We'll have to investigate whether changing the behavior will cause problems for code that uses httpuv. I believe that shiny and plumber should be OK because they remove a leading ? only if it's present.

https://github.com/rstudio/shiny/blob/353615da897bb6015ca805ae3c830324c1dad95f/R/utils.R#L557-L558
https://github.com/trestletech/plumber/blob/b70993e8743ee5a8c279f532eb768e637ec13e8b/R/query-string.R#L17-L19

[dernst]

• webutils (used by opencpu) removes the leading ? if there is one (should be ok)
  https://github.com/jeroen/webutils/blob/master/R/parse_query.R#L21
• reqres (used by fiery) also checks if there's a leading ? (should be ok)
  https://github.com/thomasp85/reqres/blob/master/R/request.R#L189
• Rook's parser assumes there is no ?.

shiny/plumber seem ok to me as well. That's the R packages I found so far that do their own query string parsing.

[wch]

OK, I think we can try taking it out. We'll want to make the change after the next release of httpuv, though.