Change Uri.Path to be own type #3221

hamnis · 2020-02-25T21:08:56Z

This changes Uri.Path to be an actual class to encapsulate Path segments.
We have two boolean flags here. These can be removed by transforming
the Path type to an own ADT hierarchy.

The main motivation for this change is to represent paths as something more than a String.

I have used scala.collection.immutable.Vector instead of List as Vector allows O(1)
appends.

This change is binary incompatible.

Typeclasses:

Eq
Semigroup
Adding a monoid is possible, but there are two zero cases. Changing the Path type to an ADT will resolve that.

We have a few convenince functions here:

startWith to be able to use in the Routers
indexOf and splitAt to allow us to replace the caret stuff.
normalize: drops empty segments
toAbsolute and toRelative
merge: using RFC3986 rules
concat

Unrelated, add out to .gitignore to be able to use bloop from intellij

Representing Encoded Strings as just Strings are not really useful, so we have added a Segment class to have encoded values in.

Use uri and path string interpolation in tests

Request changes:

scriptName and pathInfo changed to be Uri.Path

rossabaker

Thanks. I think this is headed in the right direction. I have more questions than answers.

core/src/main/scala/org/http4s/Uri.scala

rossabaker · 2020-03-09T02:36:36Z

core/src/main/scala/org/http4s/Uri.scala

+  final class Path(
+      val segments: Chain[String],
+      val absolute: Boolean = false,
+      val endsWithSlash: Boolean = false)


Seems to me you could get away with an empty segment at the beginning to represent absolute, and an empty segment at the end to represent endsWithSlash, and not need three variables. But maybe that complicates other logic.

Your idea of the ADT would move toward being able to represent absolute URIs at the type level, and could help solve the problem in #3210. But it would need to carry forward to Uri. Probably complicates things too much. But ... maybe?

core/src/main/scala/org/http4s/syntax/StringSyntax.scala

rossabaker

As we discussed on Gitter, storing encoded Strings is definitely not the right model. We should internally store a newtype of encoded strings, or store decoded strings and encode them on the fly. I'm leaning toward the latter option, but I think both are defensible.

core/src/main/scala/org/http4s/Uri.scala

tests/src/test/scala/org/http4s/MessageSpec.scala

rossabaker

Looking good. Noticed a couple more nits, but I think we're about ready finally.

core/src/main/scala/org/http4s/Uri.scala

rossabaker · 2020-05-08T04:17:38Z

core/src/main/scala/org/http4s/Uri.scala

+        )
+    }
+
+    implicit val eq: Eq[Path] = Eq.fromUniversalEquals[Path]


A lot more are possible here. Hash, Order. I think Monoid. LowerBounded, if you want to get exotic. But it's more important to get this landed, and we can circle back to those.

Not sure if this semigroup passes all the laws. I guess a test is in order

I think you've even got a Monoid with empty.

We should decide if we want this to be an Order or not. It could be. I chose not to provide an Order[Method] in my recent reforms there, because I didn't know that alphabetical order was compelling. It's perhaps a bit more useful to provide a stable ordering of URIs, of which this would be a part.

There are two monoids, both Uri.Path.empty and Uri.Path.Root
Although you might simulate the Root one with adding a toAbsolute after and vise versa with toRelative

Oh. Hmm. I was thinking of empty as being the empty. Maybe we can sidestep the ambiguity by just providing Semigroup? I would need to think through the cases a bit more deeply.

core/src/main/scala/org/http4s/Uri.scala

core/src/main/scala/org/http4s/Message.scala

hamnis · 2020-05-16T12:57:39Z

core/src/main/scala/org/http4s/Uri.scala


  /**
    * Adds the path exactly as described. Any path element must be urlencoded ahead of time.
    * @param path the path string to replace
    */
-  def withPath(path: Path): Uri = copy(path = path)
+  @deprecated("Use {withPath(Uri.Path)} instead", "1.0.0-M1")


since should be checked by someone

Does that {} syntax do anything in a deprecation warning? I sometimes backtick them. Doesn't really matter. Mostly asking for my own curiosity.

I`m not sure. For some reason i was thinking scaladoc syntax here. I can change it to backticks if that is what we have used elsewhere

hamnis · 2020-05-16T12:57:39Z

core/src/main/scala/org/http4s/Uri.scala

+        )
+    }
+
+    implicit val eq: Eq[Path] = Eq.fromUniversalEquals[Path]


Not sure if this semigroup passes all the laws. I guess a test is in order

rossabaker

I think it's ready as soon as we have a test for the instances.

rossabaker · 2020-05-18T04:14:51Z

core/src/main/scala/org/http4s/Uri.scala


  /**
    * Adds the path exactly as described. Any path element must be urlencoded ahead of time.
    * @param path the path string to replace
    */
-  def withPath(path: Path): Uri = copy(path = path)
+  @deprecated("Use {withPath(Uri.Path)} instead", "1.0.0-M1")


Does that {} syntax do anything in a deprecation warning? I sometimes backtick them. Doesn't really matter. Mostly asking for my own curiosity.

core/src/main/scala/org/http4s/Uri.scala

rossabaker · 2020-05-18T04:14:51Z

core/src/main/scala/org/http4s/Uri.scala

+        )
+    }
+
+    implicit val eq: Eq[Path] = Eq.fromUniversalEquals[Path]


I think you've even got a Monoid with empty.

We should decide if we want this to be an Order or not. It could be. I chose not to provide an Order[Method] in my recent reforms there, because I didn't know that alphabetical order was compelling. It's perhaps a bit more useful to provide a stable ordering of URIs, of which this would be a part.

hamnis · 2020-05-18T08:15:45Z

How should we handle 991d59b ? scalafix or equals hack?

rossabaker · 2020-05-18T16:12:27Z

Is the issue in 991d59b that code is calling path.equals(string), and it worked when it was a type alias and was string.equals(string)? So it'll be a silent failure? Damned universal equality.

I think an asymmetric equals, which violates both the laws of Scala and Java, would be more problematic in the long run. We could perhaps log a warning in the equals method, but that's still a runtime solution. I think a scalafix lint rule (or fix) is the best way forward.

rossabaker · 2020-05-18T16:16:19Z

If it weren't Uri.Path, but just Path, would the situation be better? We wouldn't be changing the equality of an existing type, but people are just going to change their imports and not think about the ramifications. That probably doesn't save us?

hamnis · 2020-05-18T16:24:11Z

that would not help us :( Universal equality are biting us here, and it still would with a new type.
Logging on String would make it visible at least, even though runtime...

I agree that we want a scalafix, and I also think that it should be a different PR for that.

rossabaker

I'm happy with this. We can decide on a Monoid.empty and add a scalafix separately.

hamnis · 2020-05-19T09:32:20Z

I would like a second approval on this.

rossabaker · 2020-05-19T15:58:01Z

Sorry, I just gave you a merge conflict. And will probably be giving you another, as I remove that internal implementation of hashLower from 1.0.

This changes Uri.Path to be an actual class to encapsulate Path segments. We have two boolean flags here. These can be removed by transforming the `Path` type to an own `ADT` hierarchy. The main motivation for this change is to represent paths as something more than a String. I have used `scala.collection.immutable.Vector` instead of `List` as `Vector` allows O(1) appends. This change is binary incompatible. Typeclasses: * Eq * Semigroup * Adding a monoid is possible, but there are two zero cases. Changing the Path type to an ADT will resolve that. We have a few convenince functions here: * startWith to be able to use in the Routers * indexOf and splitAt to allow us to replace the caret stuff. * normalize: drops empty segments * toAbsolute and toRelative * merge: using RFC3986 rules * concat Unrelated, add `out` to .gitignore to be able to use bloop from intellij Representing Encoded Strings as just Strings are not really useful, so we have added a Segment class to have encoded values in. * Use uri and path string interpolation in tests Request changes: * scriptName and pathInfo changed to be Uri.Path

hamnis requested review from rossabaker and ChristopherDavenport Feb 25, 2020

hamnis force-pushed the uri-path-segments branch from 5b4a8e4 to d12afcb Compare Feb 25, 2020

rossabaker reviewed Mar 9, 2020

View changes

rossabaker mentioned this pull request Mar 9, 2020

Pre-signed AWS URIs expect no decoding of %2F ('/') #3233

Closed

hamnis force-pushed the uri-path-segments branch 4 times, most recently from 2812792 to 43435f3 Compare Mar 12, 2020

hamnis added the breaking Change breaks compatibility (binary or source) label Apr 28, 2020

hamnis force-pushed the uri-path-segments branch from 43435f3 to 040631d Compare Apr 29, 2020

rossabaker reviewed Apr 30, 2020

View changes

core/src/main/scala/org/http4s/Uri.scala Outdated Show resolved Hide resolved

core/src/main/scala/org/http4s/Uri.scala Outdated Show resolved Hide resolved

core/src/main/scala/org/http4s/Uri.scala Outdated Show resolved Hide resolved

hamnis force-pushed the uri-path-segments branch from 9e6d079 to 31be90a Compare Apr 30, 2020

rossabaker reviewed May 1, 2020

View changes

core/src/main/scala/org/http4s/Uri.scala Outdated Show resolved Hide resolved

rossabaker mentioned this pull request May 1, 2020

Slashes (/) are not URL-encoded in Uri string representation in query parameters #2445

Closed

hamnis force-pushed the uri-path-segments branch 2 times, most recently from c226d5e to d30f06f Compare May 1, 2020

rossabaker reviewed May 2, 2020

View changes

tests/src/test/scala/org/http4s/MessageSpec.scala Outdated Show resolved Hide resolved

hamnis force-pushed the uri-path-segments branch from 35a1489 to ec99d5c Compare May 5, 2020

rossabaker reviewed May 8, 2020

View changes

hamnis mentioned this pull request May 16, 2020

the uri string interpolator does not accept variables #3433

Open

hamnis self-assigned this May 16, 2020

hamnis force-pushed the uri-path-segments branch from 77ecca1 to 7ebd34f Compare May 16, 2020

hamnis reviewed May 16, 2020

View changes

rossabaker reviewed May 18, 2020

View changes

hamnis force-pushed the uri-path-segments branch from c74c08e to f176815 Compare May 18, 2020

hamnis marked this pull request as ready for review May 18, 2020

rossabaker approved these changes May 18, 2020

View changes

hamnis force-pushed the uri-path-segments branch from f176815 to c9331ad Compare May 19, 2020

hamnis merged commit 52c07bb into http4s:master May 25, 2020
10 checks passed

hamnis deleted the uri-path-segments branch May 25, 2020

RafalSumislawski mentioned this pull request Dec 27, 2021

Path info index fix #5793

Merged

Change Uri.Path to be own type #3221

Change Uri.Path to be own type #3221

hamnis commented Feb 25, 2020 •

edited

rossabaker left a comment

rossabaker Mar 9, 2020

rossabaker Mar 9, 2020

rossabaker left a comment

rossabaker left a comment

rossabaker May 8, 2020

hamnis May 16, 2020

rossabaker May 18, 2020

hamnis May 18, 2020

rossabaker May 18, 2020

hamnis May 16, 2020

rossabaker May 18, 2020

hamnis May 18, 2020

hamnis May 16, 2020

rossabaker left a comment

rossabaker May 18, 2020

rossabaker May 18, 2020

hamnis commented May 18, 2020

rossabaker commented May 18, 2020

rossabaker commented May 18, 2020

hamnis commented May 18, 2020

rossabaker left a comment

hamnis commented May 19, 2020

rossabaker commented May 19, 2020

Change Uri.Path to be own type #3221

Change Uri.Path to be own type #3221

Conversation

hamnis commented Feb 25, 2020 • edited

rossabaker left a comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

rossabaker left a comment

rossabaker left a comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

rossabaker left a comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

hamnis commented May 18, 2020

rossabaker commented May 18, 2020

rossabaker commented May 18, 2020

hamnis commented May 18, 2020

rossabaker left a comment

hamnis commented May 19, 2020

rossabaker commented May 19, 2020

hamnis commented Feb 25, 2020 •

edited