
### DISCLAIMER > As of 6th March 2025, debugger is still in development. We plan to merge it behind a staff-only feature flag for staff use only, followed by non-public release and then finally a public one (akin to how Git panel release was handled). This is done to ensure the best experience when it gets released. ### END OF DISCLAIMER **The current state of the debugger implementation:** https://github.com/user-attachments/assets/c4deff07-80dd-4dc6-ad2e-0c252a478fe9 https://github.com/user-attachments/assets/e1ed2345-b750-4bb6-9c97-50961b76904f ---- All the todo's are in the following channel, so it's easier to work on this together: https://zed.dev/channel/zed-debugger-11370 If you are on Linux, you can use the following command to join the channel: ```cli zed https://zed.dev/channel/zed-debugger-11370 ``` ## Current Features - Collab - Breakpoints - Sync when you (re)join a project - Sync when you add/remove a breakpoint - Sync active debug line - Stack frames - Click on stack frame - View variables that belong to the stack frame - Visit the source file - Restart stack frame (if adapter supports this) - Variables - Loaded sources - Modules - Controls - Continue - Step back - Stepping granularity (configurable) - Step into - Stepping granularity (configurable) - Step over - Stepping granularity (configurable) - Step out - Stepping granularity (configurable) - Debug console - Breakpoints - Log breakpoints - line breakpoints - Persistent between zed sessions (configurable) - Multi buffer support - Toggle disable/enable all breakpoints - Stack frames - Click on stack frame - View variables that belong to the stack frame - Visit the source file - Show collapsed stack frames - Restart stack frame (if adapter supports this) - Loaded sources - View all used loaded sources if supported by adapter. - Modules - View all used modules (if adapter supports this) - Variables - Copy value - Copy name - Copy memory reference - Set value (if adapter supports this) - keyboard navigation - Debug Console - See logs - View output that was sent from debug adapter - Output grouping - Evaluate code - Updates the variable list - Auto completion - If not supported by adapter, we will show auto-completion for existing variables - Debug Terminal - Run custom commands and change env values right inside your Zed terminal - Attach to process (if adapter supports this) - Process picker - Controls - Continue - Step back - Stepping granularity (configurable) - Step into - Stepping granularity (configurable) - Step over - Stepping granularity (configurable) - Step out - Stepping granularity (configurable) - Disconnect - Restart - Stop - Warning when a debug session exited without hitting any breakpoint - Debug view to see Adapter/RPC log messages - Testing - Fake debug adapter - Fake requests & events --- Release Notes: - N/A --------- Co-authored-by: Piotr Osiewicz <24362066+osiewicz@users.noreply.github.com> Co-authored-by: Anthony Eid <hello@anthonyeid.me> Co-authored-by: Anthony <anthony@zed.dev> Co-authored-by: Piotr Osiewicz <peterosiewicz@gmail.com> Co-authored-by: Piotr <piotr@zed.dev>
1005 lines
37 KiB
Rust
1005 lines
37 KiB
Rust
//! Project-wide storage of the tasks available, capable of updating itself from the sources set.
|
||
|
||
use std::{
|
||
borrow::Cow,
|
||
cmp::{self, Reverse},
|
||
collections::hash_map,
|
||
path::{Path, PathBuf},
|
||
sync::Arc,
|
||
};
|
||
|
||
use anyhow::{Context as _, Result};
|
||
use collections::{HashMap, HashSet, VecDeque};
|
||
use gpui::{App, AppContext as _, Entity, SharedString, Task};
|
||
use itertools::Itertools;
|
||
use language::{ContextProvider, File, Language, LanguageToolchainStore, Location};
|
||
use settings::{parse_json_with_comments, SettingsLocation, TaskKind};
|
||
use task::{
|
||
DebugTaskDefinition, ResolvedTask, TaskContext, TaskId, TaskTemplate, TaskTemplates,
|
||
TaskVariables, VariableName,
|
||
};
|
||
use text::{Point, ToPoint};
|
||
use util::{paths::PathExt as _, post_inc, NumericPrefixWithSuffix, ResultExt as _};
|
||
use worktree::WorktreeId;
|
||
|
||
use crate::worktree_store::WorktreeStore;
|
||
|
||
/// Inventory tracks available tasks for a given project.
|
||
#[derive(Debug, Default)]
|
||
pub struct Inventory {
|
||
last_scheduled_tasks: VecDeque<(TaskSourceKind, ResolvedTask)>,
|
||
templates_from_settings: ParsedTemplates,
|
||
}
|
||
|
||
#[derive(Debug, Default)]
|
||
struct ParsedTemplates {
|
||
global: Vec<TaskTemplate>,
|
||
worktree: HashMap<WorktreeId, HashMap<(Arc<Path>, TaskKind), Vec<TaskTemplate>>>,
|
||
}
|
||
|
||
/// Kind of a source the tasks are fetched from, used to display more source information in the UI.
|
||
#[derive(Debug, Clone, PartialEq, Eq, Hash, PartialOrd, Ord)]
|
||
pub enum TaskSourceKind {
|
||
/// bash-like commands spawned by users, not associated with any path
|
||
UserInput,
|
||
/// Tasks from the worktree's .zed/task.json
|
||
Worktree {
|
||
id: WorktreeId,
|
||
directory_in_worktree: PathBuf,
|
||
id_base: Cow<'static, str>,
|
||
},
|
||
/// ~/.config/zed/task.json - like global files with task definitions, applicable to any path
|
||
AbsPath {
|
||
id_base: Cow<'static, str>,
|
||
abs_path: PathBuf,
|
||
},
|
||
/// Languages-specific tasks coming from extensions.
|
||
Language { name: SharedString },
|
||
}
|
||
|
||
/// A collection of task contexts, derived from the current state of the workspace.
|
||
/// Only contains worktrees that are visible and with their root being a directory.
|
||
#[derive(Debug, Default)]
|
||
pub struct TaskContexts {
|
||
/// A context, related to the currently opened item.
|
||
/// Item can be opened from an invisible worktree, or any other, not necessarily active worktree.
|
||
pub active_item_context: Option<(Option<WorktreeId>, Option<Location>, TaskContext)>,
|
||
/// A worktree that corresponds to the active item, or the only worktree in the workspace.
|
||
pub active_worktree_context: Option<(WorktreeId, TaskContext)>,
|
||
/// If there are multiple worktrees in the workspace, all non-active ones are included here.
|
||
pub other_worktree_contexts: Vec<(WorktreeId, TaskContext)>,
|
||
}
|
||
|
||
impl TaskContexts {
|
||
pub fn active_context(&self) -> Option<&TaskContext> {
|
||
self.active_item_context
|
||
.as_ref()
|
||
.map(|(_, _, context)| context)
|
||
.or_else(|| {
|
||
self.active_worktree_context
|
||
.as_ref()
|
||
.map(|(_, context)| context)
|
||
})
|
||
}
|
||
|
||
pub fn location(&self) -> Option<&Location> {
|
||
self.active_item_context
|
||
.as_ref()
|
||
.and_then(|(_, location, _)| location.as_ref())
|
||
}
|
||
|
||
pub fn worktree(&self) -> Option<WorktreeId> {
|
||
self.active_item_context
|
||
.as_ref()
|
||
.and_then(|(worktree_id, _, _)| worktree_id.as_ref())
|
||
.or_else(|| {
|
||
self.active_worktree_context
|
||
.as_ref()
|
||
.map(|(worktree_id, _)| worktree_id)
|
||
})
|
||
.copied()
|
||
}
|
||
}
|
||
|
||
impl TaskSourceKind {
|
||
pub fn to_id_base(&self) -> String {
|
||
match self {
|
||
TaskSourceKind::UserInput => "oneshot".to_string(),
|
||
TaskSourceKind::AbsPath { id_base, abs_path } => {
|
||
format!("{id_base}_{}", abs_path.display())
|
||
}
|
||
TaskSourceKind::Worktree {
|
||
id,
|
||
id_base,
|
||
directory_in_worktree,
|
||
} => {
|
||
format!("{id_base}_{id}_{}", directory_in_worktree.display())
|
||
}
|
||
TaskSourceKind::Language { name } => format!("language_{name}"),
|
||
}
|
||
}
|
||
}
|
||
|
||
impl Inventory {
|
||
pub fn new(cx: &mut App) -> Entity<Self> {
|
||
cx.new(|_| Self::default())
|
||
}
|
||
|
||
/// Pulls its task sources relevant to the worktree and the language given,
|
||
/// returns all task templates with their source kinds, worktree tasks first, language tasks second
|
||
/// and global tasks last. No specific order inside source kinds groups.
|
||
pub fn list_tasks(
|
||
&self,
|
||
file: Option<Arc<dyn File>>,
|
||
language: Option<Arc<Language>>,
|
||
worktree: Option<WorktreeId>,
|
||
cx: &App,
|
||
) -> Vec<(TaskSourceKind, TaskTemplate)> {
|
||
let task_source_kind = language.as_ref().map(|language| TaskSourceKind::Language {
|
||
name: language.name().into(),
|
||
});
|
||
let global_tasks = self.global_templates_from_settings();
|
||
let language_tasks = language
|
||
.and_then(|language| language.context_provider()?.associated_tasks(file, cx))
|
||
.into_iter()
|
||
.flat_map(|tasks| tasks.0.into_iter())
|
||
.flat_map(|task| Some((task_source_kind.clone()?, task)))
|
||
.chain(global_tasks);
|
||
|
||
self.worktree_templates_from_settings(worktree)
|
||
.chain(language_tasks)
|
||
.collect()
|
||
}
|
||
|
||
/// Pulls its task sources relevant to the worktree and the language given and resolves them with the [`TaskContexts`] given.
|
||
/// Joins the new resolutions with the resolved tasks that were used (spawned) before,
|
||
/// orders them so that the most recently used come first, all equally used ones are ordered so that the most specific tasks come first.
|
||
/// Deduplicates the tasks by their labels and context and splits the ordered list into two: used tasks and the rest, newly resolved tasks.
|
||
pub fn used_and_current_resolved_tasks(
|
||
&self,
|
||
task_contexts: &TaskContexts,
|
||
cx: &App,
|
||
) -> (
|
||
Vec<(TaskSourceKind, ResolvedTask)>,
|
||
Vec<(TaskSourceKind, ResolvedTask)>,
|
||
) {
|
||
let worktree = task_contexts.worktree();
|
||
let location = task_contexts.location();
|
||
let language = location
|
||
.and_then(|location| location.buffer.read(cx).language_at(location.range.start));
|
||
let task_source_kind = language.as_ref().map(|language| TaskSourceKind::Language {
|
||
name: language.name().into(),
|
||
});
|
||
let file = location.and_then(|location| location.buffer.read(cx).file().cloned());
|
||
|
||
let mut task_labels_to_ids = HashMap::<String, HashSet<TaskId>>::default();
|
||
let mut lru_score = 0_u32;
|
||
let previously_spawned_tasks = self
|
||
.last_scheduled_tasks
|
||
.iter()
|
||
.rev()
|
||
.filter(|(task_kind, _)| {
|
||
if matches!(task_kind, TaskSourceKind::Language { .. }) {
|
||
Some(task_kind) == task_source_kind.as_ref()
|
||
} else {
|
||
true
|
||
}
|
||
})
|
||
.filter(|(_, resolved_task)| {
|
||
match task_labels_to_ids.entry(resolved_task.resolved_label.clone()) {
|
||
hash_map::Entry::Occupied(mut o) => {
|
||
o.get_mut().insert(resolved_task.id.clone());
|
||
// Neber allow duplicate reused tasks with the same labels
|
||
false
|
||
}
|
||
hash_map::Entry::Vacant(v) => {
|
||
v.insert(HashSet::from_iter(Some(resolved_task.id.clone())));
|
||
true
|
||
}
|
||
}
|
||
})
|
||
.map(|(task_source_kind, resolved_task)| {
|
||
(
|
||
task_source_kind.clone(),
|
||
resolved_task.clone(),
|
||
post_inc(&mut lru_score),
|
||
)
|
||
})
|
||
.sorted_unstable_by(task_lru_comparator)
|
||
.map(|(kind, task, _)| (kind, task))
|
||
.collect::<Vec<_>>();
|
||
|
||
let not_used_score = post_inc(&mut lru_score);
|
||
let global_tasks = self.global_templates_from_settings();
|
||
let language_tasks = language
|
||
.and_then(|language| language.context_provider()?.associated_tasks(file, cx))
|
||
.into_iter()
|
||
.flat_map(|tasks| tasks.0.into_iter())
|
||
.flat_map(|task| Some((task_source_kind.clone()?, task)));
|
||
let worktree_tasks = self
|
||
.worktree_templates_from_settings(worktree)
|
||
.chain(language_tasks)
|
||
.chain(global_tasks);
|
||
|
||
let new_resolved_tasks = worktree_tasks
|
||
.flat_map(|(kind, task)| {
|
||
let id_base = kind.to_id_base();
|
||
if let TaskSourceKind::Worktree { id, .. } = &kind {
|
||
None.or_else(|| {
|
||
let (_, _, item_context) = task_contexts
|
||
.active_item_context
|
||
.as_ref()
|
||
.filter(|(worktree_id, _, _)| Some(id) == worktree_id.as_ref())?;
|
||
task.resolve_task(&id_base, item_context)
|
||
})
|
||
.or_else(|| {
|
||
let (_, worktree_context) = task_contexts
|
||
.active_worktree_context
|
||
.as_ref()
|
||
.filter(|(worktree_id, _)| id == worktree_id)?;
|
||
task.resolve_task(&id_base, worktree_context)
|
||
})
|
||
.or_else(|| {
|
||
if let TaskSourceKind::Worktree { id, .. } = &kind {
|
||
let worktree_context = task_contexts
|
||
.other_worktree_contexts
|
||
.iter()
|
||
.find(|(worktree_id, _)| worktree_id == id)
|
||
.map(|(_, context)| context)?;
|
||
task.resolve_task(&id_base, worktree_context)
|
||
} else {
|
||
None
|
||
}
|
||
})
|
||
} else {
|
||
None.or_else(|| {
|
||
let (_, _, item_context) = task_contexts.active_item_context.as_ref()?;
|
||
task.resolve_task(&id_base, item_context)
|
||
})
|
||
.or_else(|| {
|
||
let (_, worktree_context) =
|
||
task_contexts.active_worktree_context.as_ref()?;
|
||
task.resolve_task(&id_base, worktree_context)
|
||
})
|
||
}
|
||
.or_else(|| task.resolve_task(&id_base, &TaskContext::default()))
|
||
.map(move |resolved_task| (kind.clone(), resolved_task, not_used_score))
|
||
})
|
||
.filter(|(_, resolved_task, _)| {
|
||
match task_labels_to_ids.entry(resolved_task.resolved_label.clone()) {
|
||
hash_map::Entry::Occupied(mut o) => {
|
||
// Allow new tasks with the same label, if their context is different
|
||
o.get_mut().insert(resolved_task.id.clone())
|
||
}
|
||
hash_map::Entry::Vacant(v) => {
|
||
v.insert(HashSet::from_iter(Some(resolved_task.id.clone())));
|
||
true
|
||
}
|
||
}
|
||
})
|
||
.sorted_unstable_by(task_lru_comparator)
|
||
.map(|(kind, task, _)| (kind, task))
|
||
.collect::<Vec<_>>();
|
||
|
||
(previously_spawned_tasks, new_resolved_tasks)
|
||
}
|
||
|
||
/// Returns the last scheduled task by task_id if provided.
|
||
/// Otherwise, returns the last scheduled task.
|
||
pub fn last_scheduled_task(
|
||
&self,
|
||
task_id: Option<&TaskId>,
|
||
) -> Option<(TaskSourceKind, ResolvedTask)> {
|
||
if let Some(task_id) = task_id {
|
||
self.last_scheduled_tasks
|
||
.iter()
|
||
.find(|(_, task)| &task.id == task_id)
|
||
.cloned()
|
||
} else {
|
||
self.last_scheduled_tasks.back().cloned()
|
||
}
|
||
}
|
||
|
||
/// Registers task "usage" as being scheduled – to be used for LRU sorting when listing all tasks.
|
||
pub fn task_scheduled(
|
||
&mut self,
|
||
task_source_kind: TaskSourceKind,
|
||
resolved_task: ResolvedTask,
|
||
) {
|
||
self.last_scheduled_tasks
|
||
.push_back((task_source_kind, resolved_task));
|
||
if self.last_scheduled_tasks.len() > 5_000 {
|
||
self.last_scheduled_tasks.pop_front();
|
||
}
|
||
}
|
||
|
||
/// Deletes a resolved task from history, using its id.
|
||
/// A similar may still resurface in `used_and_current_resolved_tasks` when its [`TaskTemplate`] is resolved again.
|
||
pub fn delete_previously_used(&mut self, id: &TaskId) {
|
||
self.last_scheduled_tasks.retain(|(_, task)| &task.id != id);
|
||
}
|
||
|
||
fn global_templates_from_settings(
|
||
&self,
|
||
) -> impl '_ + Iterator<Item = (TaskSourceKind, TaskTemplate)> {
|
||
self.templates_from_settings
|
||
.global
|
||
.clone()
|
||
.into_iter()
|
||
.map(|template| {
|
||
(
|
||
TaskSourceKind::AbsPath {
|
||
id_base: match template.task_type {
|
||
task::TaskType::Script => Cow::Borrowed("global tasks.json"),
|
||
task::TaskType::Debug(_) => Cow::Borrowed("global debug.json"),
|
||
},
|
||
abs_path: match template.task_type {
|
||
task::TaskType::Script => paths::tasks_file().clone(),
|
||
task::TaskType::Debug(_) => paths::debug_tasks_file().clone(),
|
||
},
|
||
},
|
||
template,
|
||
)
|
||
})
|
||
}
|
||
|
||
fn worktree_templates_from_settings(
|
||
&self,
|
||
worktree: Option<WorktreeId>,
|
||
) -> impl '_ + Iterator<Item = (TaskSourceKind, TaskTemplate)> {
|
||
worktree.into_iter().flat_map(|worktree| {
|
||
self.templates_from_settings
|
||
.worktree
|
||
.get(&worktree)
|
||
.into_iter()
|
||
.flatten()
|
||
.flat_map(|(directory, templates)| {
|
||
templates.iter().map(move |template| (directory, template))
|
||
})
|
||
.map(move |((directory, _task_kind), template)| {
|
||
(
|
||
TaskSourceKind::Worktree {
|
||
id: worktree,
|
||
directory_in_worktree: directory.to_path_buf(),
|
||
id_base: Cow::Owned(format!(
|
||
"local worktree tasks from directory {directory:?}"
|
||
)),
|
||
},
|
||
template.clone(),
|
||
)
|
||
})
|
||
})
|
||
}
|
||
|
||
/// Updates in-memory task metadata from the JSON string given.
|
||
/// Will fail if the JSON is not a valid array of objects, but will continue if any object will not parse into a [`TaskTemplate`].
|
||
///
|
||
/// Global tasks are updated for no worktree provided, otherwise the worktree metadata for a given path will be updated.
|
||
pub(crate) fn update_file_based_tasks(
|
||
&mut self,
|
||
location: Option<SettingsLocation<'_>>,
|
||
raw_tasks_json: Option<&str>,
|
||
task_kind: TaskKind,
|
||
) -> anyhow::Result<()> {
|
||
let raw_tasks =
|
||
parse_json_with_comments::<Vec<serde_json::Value>>(raw_tasks_json.unwrap_or("[]"))
|
||
.context("parsing tasks file content as a JSON array")?;
|
||
let new_templates = raw_tasks
|
||
.into_iter()
|
||
.filter_map(|raw_template| match &task_kind {
|
||
TaskKind::Script => serde_json::from_value::<TaskTemplate>(raw_template).log_err(),
|
||
TaskKind::Debug => serde_json::from_value::<DebugTaskDefinition>(raw_template)
|
||
.log_err()
|
||
.and_then(|content| content.to_zed_format().log_err()),
|
||
});
|
||
|
||
let parsed_templates = &mut self.templates_from_settings;
|
||
match location {
|
||
Some(location) => {
|
||
let new_templates = new_templates.collect::<Vec<_>>();
|
||
if new_templates.is_empty() {
|
||
if let Some(worktree_tasks) =
|
||
parsed_templates.worktree.get_mut(&location.worktree_id)
|
||
{
|
||
worktree_tasks.remove(&(Arc::from(location.path), task_kind));
|
||
}
|
||
} else {
|
||
parsed_templates
|
||
.worktree
|
||
.entry(location.worktree_id)
|
||
.or_default()
|
||
.insert((Arc::from(location.path), task_kind), new_templates);
|
||
}
|
||
}
|
||
None => parsed_templates.global = new_templates.collect(),
|
||
}
|
||
Ok(())
|
||
}
|
||
}
|
||
|
||
fn task_lru_comparator(
|
||
(kind_a, task_a, lru_score_a): &(TaskSourceKind, ResolvedTask, u32),
|
||
(kind_b, task_b, lru_score_b): &(TaskSourceKind, ResolvedTask, u32),
|
||
) -> cmp::Ordering {
|
||
lru_score_a
|
||
// First, display recently used templates above all.
|
||
.cmp(lru_score_b)
|
||
// Then, ensure more specific sources are displayed first.
|
||
.then(task_source_kind_preference(kind_a).cmp(&task_source_kind_preference(kind_b)))
|
||
// After that, display first more specific tasks, using more template variables.
|
||
// Bonus points for tasks with symbol variables.
|
||
.then(task_variables_preference(task_a).cmp(&task_variables_preference(task_b)))
|
||
// Finally, sort by the resolved label, but a bit more specifically, to avoid mixing letters and digits.
|
||
.then({
|
||
NumericPrefixWithSuffix::from_numeric_prefixed_str(&task_a.resolved_label)
|
||
.cmp(&NumericPrefixWithSuffix::from_numeric_prefixed_str(
|
||
&task_b.resolved_label,
|
||
))
|
||
.then(task_a.resolved_label.cmp(&task_b.resolved_label))
|
||
.then(kind_a.cmp(kind_b))
|
||
})
|
||
}
|
||
|
||
fn task_source_kind_preference(kind: &TaskSourceKind) -> u32 {
|
||
match kind {
|
||
TaskSourceKind::Language { .. } => 1,
|
||
TaskSourceKind::UserInput => 2,
|
||
TaskSourceKind::Worktree { .. } => 3,
|
||
TaskSourceKind::AbsPath { .. } => 4,
|
||
}
|
||
}
|
||
|
||
fn task_variables_preference(task: &ResolvedTask) -> Reverse<usize> {
|
||
let task_variables = task.substituted_variables();
|
||
Reverse(if task_variables.contains(&VariableName::Symbol) {
|
||
task_variables.len() + 1
|
||
} else {
|
||
task_variables.len()
|
||
})
|
||
}
|
||
|
||
#[cfg(test)]
|
||
mod test_inventory {
|
||
use gpui::{Entity, TestAppContext};
|
||
use itertools::Itertools;
|
||
use task::TaskContext;
|
||
use worktree::WorktreeId;
|
||
|
||
use crate::Inventory;
|
||
|
||
use super::TaskSourceKind;
|
||
|
||
pub(super) fn task_template_names(
|
||
inventory: &Entity<Inventory>,
|
||
worktree: Option<WorktreeId>,
|
||
cx: &mut TestAppContext,
|
||
) -> Vec<String> {
|
||
inventory.update(cx, |inventory, cx| {
|
||
inventory
|
||
.list_tasks(None, None, worktree, cx)
|
||
.into_iter()
|
||
.map(|(_, task)| task.label)
|
||
.sorted()
|
||
.collect()
|
||
})
|
||
}
|
||
|
||
pub(super) fn register_task_used(
|
||
inventory: &Entity<Inventory>,
|
||
task_name: &str,
|
||
cx: &mut TestAppContext,
|
||
) {
|
||
inventory.update(cx, |inventory, cx| {
|
||
let (task_source_kind, task) = inventory
|
||
.list_tasks(None, None, None, cx)
|
||
.into_iter()
|
||
.find(|(_, task)| task.label == task_name)
|
||
.unwrap_or_else(|| panic!("Failed to find task with name {task_name}"));
|
||
let id_base = task_source_kind.to_id_base();
|
||
inventory.task_scheduled(
|
||
task_source_kind.clone(),
|
||
task.resolve_task(&id_base, &TaskContext::default())
|
||
.unwrap_or_else(|| panic!("Failed to resolve task with name {task_name}")),
|
||
);
|
||
});
|
||
}
|
||
|
||
pub(super) async fn list_tasks(
|
||
inventory: &Entity<Inventory>,
|
||
worktree: Option<WorktreeId>,
|
||
cx: &mut TestAppContext,
|
||
) -> Vec<(TaskSourceKind, String)> {
|
||
inventory.update(cx, |inventory, cx| {
|
||
let task_context = &TaskContext::default();
|
||
inventory
|
||
.list_tasks(None, None, worktree, cx)
|
||
.into_iter()
|
||
.filter_map(|(source_kind, task)| {
|
||
let id_base = source_kind.to_id_base();
|
||
Some((source_kind, task.resolve_task(&id_base, task_context)?))
|
||
})
|
||
.map(|(source_kind, resolved_task)| (source_kind, resolved_task.resolved_label))
|
||
.collect()
|
||
})
|
||
}
|
||
}
|
||
|
||
/// A context provided that tries to provide values for all non-custom [`VariableName`] variants for a currently opened file.
|
||
/// Applied as a base for every custom [`ContextProvider`] unless explicitly oped out.
|
||
pub struct BasicContextProvider {
|
||
worktree_store: Entity<WorktreeStore>,
|
||
}
|
||
|
||
impl BasicContextProvider {
|
||
pub fn new(worktree_store: Entity<WorktreeStore>) -> Self {
|
||
Self { worktree_store }
|
||
}
|
||
}
|
||
impl ContextProvider for BasicContextProvider {
|
||
fn build_context(
|
||
&self,
|
||
_: &TaskVariables,
|
||
location: &Location,
|
||
_: Option<HashMap<String, String>>,
|
||
_: Arc<dyn LanguageToolchainStore>,
|
||
cx: &mut App,
|
||
) -> Task<Result<TaskVariables>> {
|
||
let buffer = location.buffer.read(cx);
|
||
let buffer_snapshot = buffer.snapshot();
|
||
let symbols = buffer_snapshot.symbols_containing(location.range.start, None);
|
||
let symbol = symbols.unwrap_or_default().last().map(|symbol| {
|
||
let range = symbol
|
||
.name_ranges
|
||
.last()
|
||
.cloned()
|
||
.unwrap_or(0..symbol.text.len());
|
||
symbol.text[range].to_string()
|
||
});
|
||
|
||
let current_file = buffer
|
||
.file()
|
||
.and_then(|file| file.as_local())
|
||
.map(|file| file.abs_path(cx).to_sanitized_string());
|
||
let Point { row, column } = location.range.start.to_point(&buffer_snapshot);
|
||
let row = row + 1;
|
||
let column = column + 1;
|
||
let selected_text = buffer
|
||
.chars_for_range(location.range.clone())
|
||
.collect::<String>();
|
||
|
||
let mut task_variables = TaskVariables::from_iter([
|
||
(VariableName::Row, row.to_string()),
|
||
(VariableName::Column, column.to_string()),
|
||
]);
|
||
|
||
if let Some(symbol) = symbol {
|
||
task_variables.insert(VariableName::Symbol, symbol);
|
||
}
|
||
if !selected_text.trim().is_empty() {
|
||
task_variables.insert(VariableName::SelectedText, selected_text);
|
||
}
|
||
let worktree_root_dir =
|
||
buffer
|
||
.file()
|
||
.map(|file| file.worktree_id(cx))
|
||
.and_then(|worktree_id| {
|
||
self.worktree_store
|
||
.read(cx)
|
||
.worktree_for_id(worktree_id, cx)
|
||
.and_then(|worktree| worktree.read(cx).root_dir())
|
||
});
|
||
if let Some(worktree_path) = worktree_root_dir {
|
||
task_variables.insert(
|
||
VariableName::WorktreeRoot,
|
||
worktree_path.to_sanitized_string(),
|
||
);
|
||
if let Some(full_path) = current_file.as_ref() {
|
||
let relative_path = pathdiff::diff_paths(full_path, worktree_path);
|
||
if let Some(relative_path) = relative_path {
|
||
task_variables.insert(
|
||
VariableName::RelativeFile,
|
||
relative_path.to_sanitized_string(),
|
||
);
|
||
}
|
||
}
|
||
}
|
||
|
||
if let Some(path_as_string) = current_file {
|
||
let path = Path::new(&path_as_string);
|
||
if let Some(filename) = path.file_name().and_then(|f| f.to_str()) {
|
||
task_variables.insert(VariableName::Filename, String::from(filename));
|
||
}
|
||
|
||
if let Some(stem) = path.file_stem().and_then(|s| s.to_str()) {
|
||
task_variables.insert(VariableName::Stem, stem.into());
|
||
}
|
||
|
||
if let Some(dirname) = path.parent().and_then(|s| s.to_str()) {
|
||
task_variables.insert(VariableName::Dirname, dirname.into());
|
||
}
|
||
|
||
task_variables.insert(VariableName::File, path_as_string);
|
||
}
|
||
|
||
Task::ready(Ok(task_variables))
|
||
}
|
||
}
|
||
|
||
/// A ContextProvider that doesn't provide any task variables on it's own, though it has some associated tasks.
|
||
pub struct ContextProviderWithTasks {
|
||
templates: TaskTemplates,
|
||
}
|
||
|
||
impl ContextProviderWithTasks {
|
||
pub fn new(definitions: TaskTemplates) -> Self {
|
||
Self {
|
||
templates: definitions,
|
||
}
|
||
}
|
||
}
|
||
|
||
impl ContextProvider for ContextProviderWithTasks {
|
||
fn associated_tasks(
|
||
&self,
|
||
_: Option<Arc<dyn language::File>>,
|
||
_: &App,
|
||
) -> Option<TaskTemplates> {
|
||
Some(self.templates.clone())
|
||
}
|
||
}
|
||
|
||
#[cfg(test)]
|
||
mod tests {
|
||
use gpui::TestAppContext;
|
||
use pretty_assertions::assert_eq;
|
||
use serde_json::json;
|
||
|
||
use crate::task_store::TaskStore;
|
||
|
||
use super::test_inventory::*;
|
||
use super::*;
|
||
|
||
#[gpui::test]
|
||
async fn test_task_list_sorting(cx: &mut TestAppContext) {
|
||
init_test(cx);
|
||
let inventory = cx.update(Inventory::new);
|
||
let initial_tasks = resolved_task_names(&inventory, None, cx).await;
|
||
assert!(
|
||
initial_tasks.is_empty(),
|
||
"No tasks expected for empty inventory, but got {initial_tasks:?}"
|
||
);
|
||
let initial_tasks = task_template_names(&inventory, None, cx);
|
||
assert!(
|
||
initial_tasks.is_empty(),
|
||
"No tasks expected for empty inventory, but got {initial_tasks:?}"
|
||
);
|
||
cx.run_until_parked();
|
||
let expected_initial_state = [
|
||
"1_a_task".to_string(),
|
||
"1_task".to_string(),
|
||
"2_task".to_string(),
|
||
"3_task".to_string(),
|
||
];
|
||
|
||
inventory.update(cx, |inventory, _| {
|
||
inventory
|
||
.update_file_based_tasks(
|
||
None,
|
||
Some(&mock_tasks_from_names(
|
||
expected_initial_state.iter().map(|name| name.as_str()),
|
||
)),
|
||
settings::TaskKind::Script,
|
||
)
|
||
.unwrap();
|
||
});
|
||
assert_eq!(
|
||
task_template_names(&inventory, None, cx),
|
||
&expected_initial_state,
|
||
);
|
||
assert_eq!(
|
||
resolved_task_names(&inventory, None, cx).await,
|
||
&expected_initial_state,
|
||
"Tasks with equal amount of usages should be sorted alphanumerically"
|
||
);
|
||
|
||
register_task_used(&inventory, "2_task", cx);
|
||
assert_eq!(
|
||
task_template_names(&inventory, None, cx),
|
||
&expected_initial_state,
|
||
);
|
||
assert_eq!(
|
||
resolved_task_names(&inventory, None, cx).await,
|
||
vec![
|
||
"2_task".to_string(),
|
||
"1_a_task".to_string(),
|
||
"1_task".to_string(),
|
||
"3_task".to_string()
|
||
],
|
||
);
|
||
|
||
register_task_used(&inventory, "1_task", cx);
|
||
register_task_used(&inventory, "1_task", cx);
|
||
register_task_used(&inventory, "1_task", cx);
|
||
register_task_used(&inventory, "3_task", cx);
|
||
assert_eq!(
|
||
task_template_names(&inventory, None, cx),
|
||
&expected_initial_state,
|
||
);
|
||
assert_eq!(
|
||
resolved_task_names(&inventory, None, cx).await,
|
||
vec![
|
||
"3_task".to_string(),
|
||
"1_task".to_string(),
|
||
"2_task".to_string(),
|
||
"1_a_task".to_string(),
|
||
],
|
||
);
|
||
|
||
inventory.update(cx, |inventory, _| {
|
||
inventory
|
||
.update_file_based_tasks(
|
||
None,
|
||
Some(&mock_tasks_from_names(
|
||
["10_hello", "11_hello"]
|
||
.into_iter()
|
||
.chain(expected_initial_state.iter().map(|name| name.as_str())),
|
||
)),
|
||
settings::TaskKind::Script,
|
||
)
|
||
.unwrap();
|
||
});
|
||
cx.run_until_parked();
|
||
let expected_updated_state = [
|
||
"10_hello".to_string(),
|
||
"11_hello".to_string(),
|
||
"1_a_task".to_string(),
|
||
"1_task".to_string(),
|
||
"2_task".to_string(),
|
||
"3_task".to_string(),
|
||
];
|
||
assert_eq!(
|
||
task_template_names(&inventory, None, cx),
|
||
&expected_updated_state,
|
||
);
|
||
assert_eq!(
|
||
resolved_task_names(&inventory, None, cx).await,
|
||
vec![
|
||
"3_task".to_string(),
|
||
"1_task".to_string(),
|
||
"2_task".to_string(),
|
||
"1_a_task".to_string(),
|
||
"10_hello".to_string(),
|
||
"11_hello".to_string(),
|
||
],
|
||
);
|
||
|
||
register_task_used(&inventory, "11_hello", cx);
|
||
assert_eq!(
|
||
task_template_names(&inventory, None, cx),
|
||
&expected_updated_state,
|
||
);
|
||
assert_eq!(
|
||
resolved_task_names(&inventory, None, cx).await,
|
||
vec![
|
||
"11_hello".to_string(),
|
||
"3_task".to_string(),
|
||
"1_task".to_string(),
|
||
"2_task".to_string(),
|
||
"1_a_task".to_string(),
|
||
"10_hello".to_string(),
|
||
],
|
||
);
|
||
}
|
||
|
||
#[gpui::test]
|
||
async fn test_inventory_static_task_filters(cx: &mut TestAppContext) {
|
||
init_test(cx);
|
||
let inventory = cx.update(Inventory::new);
|
||
let common_name = "common_task_name";
|
||
let worktree_1 = WorktreeId::from_usize(1);
|
||
let worktree_2 = WorktreeId::from_usize(2);
|
||
|
||
cx.run_until_parked();
|
||
let worktree_independent_tasks = vec![
|
||
(
|
||
TaskSourceKind::AbsPath {
|
||
id_base: "global tasks.json".into(),
|
||
abs_path: paths::tasks_file().clone(),
|
||
},
|
||
common_name.to_string(),
|
||
),
|
||
(
|
||
TaskSourceKind::AbsPath {
|
||
id_base: "global tasks.json".into(),
|
||
abs_path: paths::tasks_file().clone(),
|
||
},
|
||
"static_source_1".to_string(),
|
||
),
|
||
(
|
||
TaskSourceKind::AbsPath {
|
||
id_base: "global tasks.json".into(),
|
||
abs_path: paths::tasks_file().clone(),
|
||
},
|
||
"static_source_2".to_string(),
|
||
),
|
||
];
|
||
let worktree_1_tasks = [
|
||
(
|
||
TaskSourceKind::Worktree {
|
||
id: worktree_1,
|
||
directory_in_worktree: PathBuf::from(".zed"),
|
||
id_base: "local worktree tasks from directory \".zed\"".into(),
|
||
},
|
||
common_name.to_string(),
|
||
),
|
||
(
|
||
TaskSourceKind::Worktree {
|
||
id: worktree_1,
|
||
directory_in_worktree: PathBuf::from(".zed"),
|
||
id_base: "local worktree tasks from directory \".zed\"".into(),
|
||
},
|
||
"worktree_1".to_string(),
|
||
),
|
||
];
|
||
let worktree_2_tasks = [
|
||
(
|
||
TaskSourceKind::Worktree {
|
||
id: worktree_2,
|
||
directory_in_worktree: PathBuf::from(".zed"),
|
||
id_base: "local worktree tasks from directory \".zed\"".into(),
|
||
},
|
||
common_name.to_string(),
|
||
),
|
||
(
|
||
TaskSourceKind::Worktree {
|
||
id: worktree_2,
|
||
directory_in_worktree: PathBuf::from(".zed"),
|
||
id_base: "local worktree tasks from directory \".zed\"".into(),
|
||
},
|
||
"worktree_2".to_string(),
|
||
),
|
||
];
|
||
|
||
inventory.update(cx, |inventory, _| {
|
||
inventory
|
||
.update_file_based_tasks(
|
||
None,
|
||
Some(&mock_tasks_from_names(
|
||
worktree_independent_tasks
|
||
.iter()
|
||
.map(|(_, name)| name.as_str()),
|
||
)),
|
||
settings::TaskKind::Script,
|
||
)
|
||
.unwrap();
|
||
inventory
|
||
.update_file_based_tasks(
|
||
Some(SettingsLocation {
|
||
worktree_id: worktree_1,
|
||
path: Path::new(".zed"),
|
||
}),
|
||
Some(&mock_tasks_from_names(
|
||
worktree_1_tasks.iter().map(|(_, name)| name.as_str()),
|
||
)),
|
||
settings::TaskKind::Script,
|
||
)
|
||
.unwrap();
|
||
inventory
|
||
.update_file_based_tasks(
|
||
Some(SettingsLocation {
|
||
worktree_id: worktree_2,
|
||
path: Path::new(".zed"),
|
||
}),
|
||
Some(&mock_tasks_from_names(
|
||
worktree_2_tasks.iter().map(|(_, name)| name.as_str()),
|
||
)),
|
||
settings::TaskKind::Script,
|
||
)
|
||
.unwrap();
|
||
});
|
||
|
||
assert_eq!(
|
||
list_tasks_sorted_by_last_used(&inventory, None, cx).await,
|
||
worktree_independent_tasks,
|
||
"Without a worktree, only worktree-independent tasks should be listed"
|
||
);
|
||
assert_eq!(
|
||
list_tasks_sorted_by_last_used(&inventory, Some(worktree_1), cx).await,
|
||
worktree_1_tasks
|
||
.iter()
|
||
.chain(worktree_independent_tasks.iter())
|
||
.cloned()
|
||
.sorted_by_key(|(kind, label)| (task_source_kind_preference(kind), label.clone()))
|
||
.collect::<Vec<_>>(),
|
||
);
|
||
assert_eq!(
|
||
list_tasks_sorted_by_last_used(&inventory, Some(worktree_2), cx).await,
|
||
worktree_2_tasks
|
||
.iter()
|
||
.chain(worktree_independent_tasks.iter())
|
||
.cloned()
|
||
.sorted_by_key(|(kind, label)| (task_source_kind_preference(kind), label.clone()))
|
||
.collect::<Vec<_>>(),
|
||
);
|
||
|
||
assert_eq!(
|
||
list_tasks(&inventory, None, cx).await,
|
||
worktree_independent_tasks,
|
||
"Without a worktree, only worktree-independent tasks should be listed"
|
||
);
|
||
assert_eq!(
|
||
list_tasks(&inventory, Some(worktree_1), cx).await,
|
||
worktree_1_tasks
|
||
.iter()
|
||
.chain(worktree_independent_tasks.iter())
|
||
.cloned()
|
||
.collect::<Vec<_>>(),
|
||
);
|
||
assert_eq!(
|
||
list_tasks(&inventory, Some(worktree_2), cx).await,
|
||
worktree_2_tasks
|
||
.iter()
|
||
.chain(worktree_independent_tasks.iter())
|
||
.cloned()
|
||
.collect::<Vec<_>>(),
|
||
);
|
||
}
|
||
|
||
fn init_test(_cx: &mut TestAppContext) {
|
||
if std::env::var("RUST_LOG").is_ok() {
|
||
env_logger::try_init().ok();
|
||
}
|
||
TaskStore::init(None);
|
||
}
|
||
|
||
async fn resolved_task_names(
|
||
inventory: &Entity<Inventory>,
|
||
worktree: Option<WorktreeId>,
|
||
cx: &mut TestAppContext,
|
||
) -> Vec<String> {
|
||
let (used, current) = inventory.update(cx, |inventory, cx| {
|
||
let mut task_contexts = TaskContexts::default();
|
||
task_contexts.active_worktree_context =
|
||
worktree.map(|worktree| (worktree, TaskContext::default()));
|
||
inventory.used_and_current_resolved_tasks(&task_contexts, cx)
|
||
});
|
||
used.into_iter()
|
||
.chain(current)
|
||
.map(|(_, task)| task.original_task().label.clone())
|
||
.collect()
|
||
}
|
||
|
||
fn mock_tasks_from_names<'a>(task_names: impl Iterator<Item = &'a str> + 'a) -> String {
|
||
serde_json::to_string(&serde_json::Value::Array(
|
||
task_names
|
||
.map(|task_name| {
|
||
json!({
|
||
"label": task_name,
|
||
"command": "echo",
|
||
"args": vec![task_name],
|
||
})
|
||
})
|
||
.collect::<Vec<_>>(),
|
||
))
|
||
.unwrap()
|
||
}
|
||
|
||
async fn list_tasks_sorted_by_last_used(
|
||
inventory: &Entity<Inventory>,
|
||
worktree: Option<WorktreeId>,
|
||
cx: &mut TestAppContext,
|
||
) -> Vec<(TaskSourceKind, String)> {
|
||
let (used, current) = inventory.update(cx, |inventory, cx| {
|
||
let mut task_contexts = TaskContexts::default();
|
||
task_contexts.active_worktree_context =
|
||
worktree.map(|worktree| (worktree, TaskContext::default()));
|
||
inventory.used_and_current_resolved_tasks(&task_contexts, cx)
|
||
});
|
||
let mut all = used;
|
||
all.extend(current);
|
||
all.into_iter()
|
||
.map(|(source_kind, task)| (source_kind, task.resolved_label))
|
||
.sorted_by_key(|(kind, label)| (task_source_kind_preference(kind), label.clone()))
|
||
.collect()
|
||
}
|
||
}
|