AST-based modification of turbo.json (#5509)

When we run `link` and `unlink` we want those commands to create the
smallest set of changes possible in the output while also accounting for
edge cases like multiple definition.

This PR adds an AST-aware JSON-rewriter that modifies the file contents
rather than attempting to deserialize and reserialize the entire
document (which would otherwise be lossy unless deserialized to a
concrete syntax tree).

The only implementation of a concrete syntax tree for JSON in the Rust
ecosystem appears to be
https://github.com/rome/tools/tree/main/crates/rome_json_parser which
presently lexes `jsonc`, but does not link it into the rest of the
parser. There is an open, abandoned PR for that effort:
rome/tools#4382

After review, I decided that the additional complexity and size of
pulling in the entire Rome parsing toolchain was not going to be
worthwhile at this time for a task this small in scope.

I investigated `pest` as an alternative, but grammar correctness edge
cases for JSON are not great, and the whitespace and comment macros in
the `pest` grammar would have to be hand-implemented. I elected against
this for low confidence in a toy parser.

Eventually I chose to use `jsonc_parser` which doesn't actually create
all of the necessary tokens, but using conservative patterns on top of
the limited information I can ensure that it doesn't _fail_. However,
this greatly limits the formatting abilities that we can accomplish. I
decided this was an acceptable tradeoff for complexity, total effort,
dep size, scope of use (just `link` and `unlink` setting one property).

Additional output formatting improvements are out-of-scope for this PR.
This generates assuredly parseable content, but makes no guarantees as
to what the output format looks like.

---------

Co-authored-by: Nathan Hammond <Nathan Hammond>

crates/turborepo-lib/Cargo.toml

@@ -56,6 +56,7 @@ humantime = "2.1.0"
 indicatif = { workspace = true }
 itertools = { workspace = true }
 json_comments = "0.2.1"
+jsonc-parser = { version = "0.21.0" }
 lazy_static = { workspace = true }
 libc = "0.2.140"
 notify = "5.1"

crates/turborepo-lib/src/commands/link.rs

@@ -25,6 +25,7 @@ use crate::{
     cli::LinkTarget,
     commands::CommandBase,
     config::{RawTurboJSON, SpacesJson},
+    rewrite_json,
 };
 
 #[derive(Clone)]
@@ -443,29 +444,16 @@ fn add_turbo_to_gitignore(base: &CommandBase) -> Result<()> {
 
 fn add_space_id_to_turbo_json(base: &CommandBase, space_id: &str) -> Result<()> {
     let turbo_json_path = base.repo_root.join_component("turbo.json");
+    let turbo_json = fs::read_to_string(&turbo_json_path)?;
+    let space_id_json_value = format!("\"{}\"", space_id);
 
-    if !turbo_json_path.exists() {
-        return Err(anyhow!("turbo.json not found."));
-    }
-
-    let turbo_json_file = File::open(&turbo_json_path)?;
-    let mut turbo_json: RawTurboJSON = serde_json::from_reader(turbo_json_file)?;
-    match turbo_json.experimental_spaces {
-        Some(mut spaces_config) => {
-            spaces_config.id = Some(space_id.to_string());
-            turbo_json.experimental_spaces = Some(spaces_config);
-        }
-        None => {
-            turbo_json.experimental_spaces = Some(SpacesJson {
-                id: Some(space_id.to_string()),
-                other: None,
-            });
-        }
-    }
+    let output = rewrite_json::set_path(
+        &turbo_json,
+        &["experimentalSpaces", "id"],
+        &space_id_json_value,
+    )?;
 
-    // write turbo_json back to file
-    let config_file = File::create(&turbo_json_path)?;
-    serde_json::to_writer_pretty(&config_file, &turbo_json)?;
+    fs::write(turbo_json_path, output)?;
 
     Ok(())
 }

crates/turborepo-lib/src/commands/unlink.rs

@@ -1,9 +1,9 @@
-use std::{fs, fs::File};
+use std::fs;
 
 use anyhow::{Context, Result};
 use turborepo_ui::GREY;
 
-use crate::{cli::LinkTarget, commands::CommandBase, config::RawTurboJSON};
+use crate::{cli::LinkTarget, commands::CommandBase, config::RawTurboJSON, rewrite_json};
 
 enum UnlinkSpacesResult {
     Unlinked,
@@ -55,25 +55,13 @@ pub fn unlink(base: &mut CommandBase, target: LinkTarget) -> Result<()> {
 
 fn remove_spaces_from_turbo_json(base: &CommandBase) -> Result<UnlinkSpacesResult> {
     let turbo_json_path = base.repo_root.join_component("turbo.json");
+    let turbo_json = fs::read_to_string(&turbo_json_path)?;
 
-    let turbo_json_contents =
-        fs::read_to_string(&turbo_json_path).context("unable to open turbo.json file")?;
-    let mut turbo_json: RawTurboJSON = serde_json::from_str(&turbo_json_contents)?;
-    let has_spaces_id = turbo_json
-        .experimental_spaces
-        .unwrap_or_default()
-        .id
-        .is_some();
-    // remove the spaces config
-    // TODO: in the future unlink should possible just remove the spaces id
-    turbo_json.experimental_spaces = None;
-
-    // write turbo_json back to file
-    let config_file = File::create(&turbo_json_path)?;
-    serde_json::to_writer_pretty(&config_file, &turbo_json)?;
-
-    match has_spaces_id {
-        true => Ok(UnlinkSpacesResult::Unlinked),
-        false => Ok(UnlinkSpacesResult::NoSpacesFound),
+    let output = rewrite_json::unset_path(&turbo_json, &["experimentalSpaces", "id"])?;
+    if let Some(output) = output {
+        fs::write(turbo_json_path, output)?;
+        Ok(UnlinkSpacesResult::Unlinked)
+    } else {
+        Ok(UnlinkSpacesResult::NoSpacesFound)
     }
 }

crates/turborepo-lib/src/lib.rs

@@ -18,6 +18,7 @@ mod opts;
 mod package_graph;
 mod package_json;
 mod package_manager;
+mod rewrite_json;
 mod run;
 mod shim;
 mod task_graph;