remove specifications of default options #185

saghm · 2017-02-03T05:56:00Z

Replaces #183 and moves us a bit closer towards #179

This is a big one; some of the changes of the options structs were relatively easy to integrate, but others (I'm looking you, FindOptions) required quite a bit of finagling with parts of the code I hadn't looked closely at in a while.

All tests pass on 3.2, and the only two failing tests on 3.4 are apm::logging (I'll need to look into this more closely) and client::db::create_and_get_users (which seems to return the users in a different order than before), both of which appear to be issues with the tests not being compatible with 3.4 rather than the driver itself. This means the driver should work with 3.4 after this is merged!

saghm · 2017-02-03T05:57:50Z

src/coll/batch.rs

@@ -10,33 +10,54 @@ pub struct DeleteModel {
    pub multi: bool,
 }

+impl DeleteModel {


I noticed that a lot of our code grouped the definitions of the types together at the top and the implementations below. This made it a bit annoying to work with (i.e. needing either two buffers or to scroll back and forth to see struct signatures at the same time as modifying the implementations), so I'm thinking that going forward we should try to group implementations with definitions.

saghm · 2017-02-03T05:59:26Z

src/coll/mod.rs

-
-        let pipeline_map = pipeline.iter()
-            .map(|bdoc| Bson::Document(bdoc.to_owned()))
+        let pipeline_map: Vec<_> = pipeline.into_iter()


I'm not sure why we were copying all of the document instead of just moving them, which seems more idiomatic since they aren't being used afterwards.

I'm sure there are many, many places were we make unnecessary copies. I look forward to destroying all of them in the future.

saghm · 2017-02-03T06:00:39Z

src/coll/mod.rs

        match result.get("n") {
-            Some(&Bson::I32(ref n)) => Ok(*n as i64),
-            Some(&Bson::I64(ref n)) => Ok(*n),
+            Some(&Bson::I32(n)) => Ok(n as i64),


Numeric types (such as i32 and i64) implement Copy, so there isn't any need to capture them by reference.

saghm · 2017-02-03T06:01:52Z

src/coll/mod.rs

-        if filter.is_some() {
-            spec.insert("query", Bson::Document(filter.unwrap()));
+
+        if let Some(filter_doc) = filter {


Although using is_some() and unwrap() is technically safe, it's more idiomatic to use if let Some(...), so I changed to this style here and a few other places.

saghm · 2017-02-03T06:03:43Z

src/coll/mod.rs

                       filter: bson::Document,
+                       options: bson::Document,
                       _max_time_ms: Option<i64>,


I've noticed we don't seem to actually use this option anywhere in the driver. I think it might be a good idea to look at how other drivers use it, since it probably is included in the spec for something.

saghm · 2017-02-03T06:28:09Z

tests/apm/mod.rs

@@ -57,7 +57,8 @@ fn logging() {
    let _ = fs::remove_file("test_apm_log.txt");

    // Reset State
-    let reset_client = Client::connect("localhost", 27017).expect("Failed to connect to localhost:27017");
+    let reset_client = Client::connect("localhost", 27017)


The most recent large rustfmt/clippy PR only checked src (and not tests), so rustfmt picked up a few things.

saghm · 2017-02-03T06:28:44Z

tests/apm/mod.rs

@@ -91,8 +92,7 @@ fn logging() {

    // Create collection started
    read_first_non_monitor_line(&mut file, &mut line);
-    assert_eq!("COMMAND.create_collection 127.0.0.1:27017 STARTED: { create: \"logging\", \
-                capped: false, auto_index_id: true, flags: 1 }\n",
+    assert_eq!("COMMAND.create_collection 127.0.0.1:27017 STARTED: { create: \"logging\" }\n",


Here (and in several places below) the logging output changed due to the default fields no longer being present.

saghm · 2017-02-03T06:30:20Z

tests/client/crud_spec/framework.rs

@@ -170,7 +170,7 @@ macro_rules! run_insert_one_test {
 macro_rules! run_replace_one_test {
    ( $db:expr, $coll:expr, $filter:expr, $replacement:expr, $upsert:expr,
        $outcome:expr ) => {{
-            let options = ReplaceOptions::new($upsert, None);
+            let options = ReplaceOptions { upsert: $upsert, write_concern: None };


I ran into a weird thing here when I tried to use the ..Default::default() syntax; the compiler seemed to be treating the .. as part of a range expression. Strangely, it only seemed to occur in certain places in macro definitions.

macros: still sketch.

Yeah, I think I probably only implemented all this in macros at the time because I had finally gotten to the point where I understood them and therefore wanted to celebrate by using them everywhere.

Let's open an issue to convert the test frameworks from macros to legitimate Rust code; it'll be a bit of work, but it'll probably save a lot of hassle in the long run.

saghm · 2017-02-03T06:32:43Z

tests/json/crud/options.rs

-            Some(Bson::Document(doc)) => Some(doc),
-            _ => None,
-        };
+        if let Some(Bson::Document(projection)) = object.get("projection").map(Bson::from_json) {


I'll be honest; I have absolutely no idea why I implemented this (and the similar cases below) in such a convoluted way originally. I'll just have to chalk it up to my lack of familiarity with Rust at that point.

saghm · 2017-02-03T06:34:25Z

src/coll/options.rs

 pub struct CountOptions {
-    pub skip: u64,
-    pub limit: i64,
+    pub skip: Option<i64>,


I'm not quite sure why an unsigned integer was used here (and in another place below). The spec doesn't use unsigned integers (probably due to there not being an easy why to enforce it in dynamically typed languages)

kyeah

the longest PR 🥇

overall, looks like a good movement towards a cleaner codebase, and one step removed from making use of derive_builder and Into<Option>. I think there are areas where we can do even less cloning (looking at you, read/write preferences), but we can look at that separately.

kyeah · 2017-02-04T22:06:02Z

src/coll/batch.rs

@@ -10,33 +10,54 @@ pub struct DeleteModel {
    pub multi: bool,
 }

+impl DeleteModel {
+    pub fn new(filter: Document, multi: bool) -> DeleteModel {


You can use #[derive(new)] now :)

https://github.com/nrc/derive-new

EDIT: Document might need to derive this, actually.

Seems like it's worth adding that in the bson crate.

kyeah · 2017-02-04T23:03:16Z

src/cursor.rs

@@ -27,7 +27,8 @@ use {Client, CommandType, Error, ErrorCode, Result, ThreadedClient};
 use apm::{CommandStarted, CommandResult, EventRunner};

 use bson::{self, Bson};
-use common::{ReadMode, ReadPreference};
+use common::{ReadMode, merge_options, ReadPreference};


tiniest nit: let's move the function import to the end (unless that's from rustfmt, in which case... 🙄)

Oops, I actually meant to put it alphabetically, which would be at the beginning. Is that okay too?

kyeah · 2017-02-04T23:06:58Z

src/cursor.rs

-                                        query.clone(),
-                                        return_field_selector);
+                                        namespace.clone(),
+                                        options.skip.unwrap_or(0) as i32,


Gonna guess it's pre-emptive specification — if they change the wire-protocol to accept i64, driver users won't have to change their code, as the API won't change. Although I recall i32s in other parts of the spec, so... 🙃

kyeah · 2017-02-07T03:45:46Z

src/coll/mod.rs

-
-        let pipeline_map = pipeline.iter()
-            .map(|bdoc| Bson::Document(bdoc.to_owned()))
+        let pipeline_map: Vec<_> = pipeline.into_iter()


I'm sure there are many, many places were we make unnecessary copies. I look forward to destroying all of them in the future.

kyeah · 2017-02-07T03:49:06Z

src/coll/mod.rs

        match result.get("n") {
-            Some(&Bson::I32(ref n)) => Ok(*n as i64),
-            Some(&Bson::I64(ref n)) => Ok(*n),
+            Some(&Bson::I32(n)) => Ok(n as i64),


kyeah · 2017-02-07T04:14:43Z

src/coll/mod.rs

-            update.insert("q", Bson::Document(model.filter));
-            update.insert("u", Bson::Document(model.update));
-            update.insert("upsert", Bson::Boolean(model.upsert));
+            let mut update = bson::Document::from(model);


kyeah · 2017-02-07T04:14:56Z

src/coll/mod.rs

+            "update" => (self.name()),
+            "updates" => updates,
+            "writeConcern" => (wc.to_bson())
+        };


kyeah · 2017-02-07T04:22:11Z

src/common.rs

@@ -95,3 +95,8 @@ impl WriteConcern {
        bson
    }
 }
+
+pub fn merge_options<T: Into<bson::Document>>(document: bson::Document, options: T) -> bson::Document {


Would this make more sense as a function on bson::Document? Feels reaal funny to have this global function, and languages like ruby have this type of method on the hash object itself. doc.merge(other_doc)

kyeah · 2017-02-07T04:26:47Z

src/db/options.rs

+
+        let mut flags = 0;
+
+        if let Some(true) = options.use_power_of_two_sizes {


oh nice! yeah, it only has two flags, so probably not worth it.

kyeah · 2017-02-07T04:28:15Z

tests/client/crud_spec/framework.rs

@@ -170,7 +170,7 @@ macro_rules! run_insert_one_test {
 macro_rules! run_replace_one_test {
    ( $db:expr, $coll:expr, $filter:expr, $replacement:expr, $upsert:expr,
        $outcome:expr ) => {{
-            let options = ReplaceOptions::new($upsert, None);
+            let options = ReplaceOptions { upsert: $upsert, write_concern: None };


macros: still sketch.

kyeah

cool; let's revisit bson crate items in a future PR.

saghm · 2017-02-12T04:15:21Z

Sounds good!

saghm commented Feb 3, 2017

View reviewed changes

kyeah self-requested a review February 4, 2017 22:42

kyeah suggested changes Feb 7, 2017

View reviewed changes

remove specifications of default options

5991a24

saghm force-pushed the remove-default-options branch from 433a2c2 to 5991a24 Compare February 11, 2017 20:12

kyeah approved these changes Feb 12, 2017

View reviewed changes

saghm merged commit cd7fe74 into mongodb-labs:master Feb 12, 2017

This was referenced Feb 12, 2017

The autoIndexId option removed in version 3.4 #183

Closed

3.4 compatibility #179

Closed

saghm deleted the remove-default-options branch April 14, 2017 05:29


		let mut flags = 0;

		if let Some(true) = options.use_power_of_two_sizes {

remove specifications of default options #185

remove specifications of default options #185

Uh oh!

Conversation

saghm commented Feb 3, 2017

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

kyeah left a comment

Choose a reason for hiding this comment

Uh oh!

kyeah Feb 4, 2017 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

kyeah left a comment

Choose a reason for hiding this comment

Uh oh!

saghm commented Feb 12, 2017

Uh oh!

Uh oh!

kyeah Feb 4, 2017 •

edited

Loading