Optimization of Device Analysis by JonatanWaern · Pull Request #217 · intel/dml-language-server

JonatanWaern · 2026-03-25T13:21:44Z

Attempts & Testing of optimization of the device analysis

Use result to do cancellation-return instead of panic
Box large enum variants
Refine obj-to-contexts mapping error propagation
Disable recursive cache
Turn off caching
Terminate message reader faster

Signed-off-by: Jonatan Waern <jonatan.waern@intel.com>

Silences internal errors from things that are not actually such, meaning we can return empty object results from contexts when we think it's fine Signed-off-by: Jonatan Waern <jonatan.waern@intel.com>

Copilot

Pull request overview

This PR focuses on optimizing device analysis execution and operational behavior by improving cancellation handling, reducing memory pressure (boxing large enum variants + boxed job/analysis payloads), and introducing bounded/truncated logging and reference-cache sizing controls.

Changes:

Introduces truncating logging macros controlled by MAX_LOG_MESSAGE_LENGTH and migrates much of the codebase to use crate::logging.
Reworks cancellation flow to return Result/AnalysisError::Cancelled instead of panicking, and shortens thread lifetimes (Weak-based message reader).
Adds a size-bounded reference cache (configurable via max_reference_cache_size) and boxes several large enum variants/analysis result payloads.

Reviewed changes

Copilot reviewed 37 out of 37 changed files in this pull request and generated 7 comments.

Show a summary per file

File	Description
USAGE.md	Documents `MAX_LOG_MESSAGE_LENGTH` environment variable.
src/vfs/mod.rs	Switches tracing to crate logging wrapper.
src/server/mod.rs	Switches to crate logging; changes some warn→info/error; message reader uses Weak to exit earlier.
src/server/message.rs	Switches to crate logging wrapper.
src/server/io.rs	Switches to crate logging wrapper.
src/server/dispatch.rs	Switches to crate logging wrapper.
src/logging.rs	Adds truncating log macros + env-controlled max message length + tests.
src/lint/mod.rs	Uses `AnalysisProcessResult` and cancellation-aware `check_alive()`.
src/file_management.rs	Switches to crate logging wrapper.
src/dfa/main.rs	Minor formatting-only diff.
src/dfa/client.rs	Switches to crate logging wrapper.
src/config.rs	Adds `max_reference_cache_size` config option (default 500MB).
src/concurrency.rs	Replaces panic-on-cancel with `AnalysisError::Cancelled` via `check_alive()`.
src/cmd.rs	Switches to crate logging wrapper.
src/analysis/templating/traits.rs	Switches to crate logging wrapper.
src/analysis/templating/topology.rs	Switches to crate logging wrapper.
src/analysis/templating/objects.rs	Boxes large variants; refactors shallow-object accessors and hooks boxing.
src/analysis/symbols.rs	Removes clippy large-enum allowance after boxing-related changes elsewhere.
src/analysis/structure/toplevel.rs	Switches to crate logging wrapper.
src/analysis/structure/statements.rs	Switches to crate logging wrapper; adapts to new `ForPre` representation.
src/analysis/structure/expressions.rs	Switches to crate logging wrapper.
src/analysis/scope.rs	Switches to crate logging wrapper.
src/analysis/parsing/structure.rs	Switches to crate logging wrapper; removes clippy large-enum allowance.
src/analysis/parsing/statement.rs	Boxes large variants (`ForPreDeclaration`, `SwitchHashIf`) and updates parsing/tests accordingly.
src/analysis/parsing/parser.rs	Switches to crate logging wrapper.
src/analysis/mod.rs	Adds `AnalysisError`/`AnalysisProcessResult`, bounded reference cache, cancellation-aware parallel reference matching, and new job options threading.
src/actions/work_pool.rs	Switches to crate logging wrapper; changes warn→error/info.
src/actions/semantic_lookup.rs	Updates symbol lookup to handle new `Result` return type.
src/actions/requests.rs	Switches to crate logging wrapper; adjusts warn→info/error.
src/actions/notifications.rs	Switches to crate logging wrapper; adjusts warn→error.
src/actions/mod.rs	Switches to crate logging wrapper; threads new `DeviceAnalysisJobOptions` into device job enqueue.
src/actions/hover.rs	Switches to crate logging wrapper (glob import).
src/actions/analysis_storage.rs	Boxes `AnalysisResult` variants; adds generic `make_timestamped` helper.
src/actions/analysis_queue.rs	Boxes queued jobs; threads job options; handles cancellation distinctly.
clients.md	Updates config.rs link target line numbers.
CHANGELOG.md	Notes new config/env controls.
Cargo.toml	Adds `hashlink` and `malloc_size_of` dependencies.

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

src/analysis/mod.rs

src/logging.rs

CHANGELOG.md

src/analysis/mod.rs

Cargo.toml

src/actions/semantic_lookup.rs

Signed-off-by: Jonatan Waern <jonatan.waern@intel.com>

This is meant to disallow infinite recursion when something breaks, but the resource overhead has proven too large (15%~ of peak mem foorprint)

Signed-off-by: Jonatan Waern <jonatan.waern@intel.com>

Copilot

Pull request overview

Copilot reviewed 37 out of 37 changed files in this pull request and generated 4 comments.

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

Copilot · 2026-04-01T14:28:59Z

CHANGELOG.md

 ## 0.9.19
-
+- Added configuration option to control the max cache size while resolving references in semantic analysis, defaulting to 500MB
+- Added environment variable to control max per-log output length for most logging from the DLS, defaulting to 1000 bytes..


The changelog entry ends with a double period ("1000 bytes.."). Please fix the punctuation to avoid shipping a typo in release notes.

Suggested change

- Added environment variable to control max per-log output length for most logging from the DLS, defaulting to 1000 bytes..

- Added environment variable to control max per-log output length for most logging from the DLS, defaulting to 1000 bytes.

Copilot · 2026-04-01T14:28:59Z

src/logging.rs

+            if let Some(max_len) = $trunc_len {
+                if max_len > 0 && s.len() > max_len {
+                    let mut slice_index = max_len;
+                    while !s.is_char_boundary(slice_index) {
+                        slice_index -= 1;
+                    }
+                    break 'check_block [&s[..slice_index], " ..."].concat();
+                }


format_truncated! appends " ..." after slicing to max_len, so the resulting log line can exceed MAX_LOG_MESSAGE_LENGTH by the suffix length. Either adjust the truncation logic so the final output length stays within the configured maximum (e.g., reserve space for the suffix), or update the variable/docs to clarify that the limit applies to the pre-suffix slice only.

Copilot · 2026-04-01T14:29:00Z

src/analysis/mod.rs

+        // Note: failing to find here is not actually an error
+        curr_objs.into_iter()
+            .map(|o|self.context_to_objs(o, context_chain, limitations))
+            .process_results(|r|r.flatten().collect())


contexts_to_objs_aux says that failing to find objects "is not actually an error", but it still propagates Err from context_to_objs (e.g., missing structural/method objects). This makes normal "not found" cases behave like hard errors and can cascade into lookup_symbols failures. Consider returning Ok(vec![]) for expected-miss cases (and reserving Err for invariants), or update the comment + callers to consistently treat these as errors.

Suggested change

// Note: failing to find here is not actually an error

curr_objs.into_iter()

.map(|o|self.context_to_objs(o, context_chain, limitations))

.process_results(|r|r.flatten().collect())

// Note: failing to find here is not actually an error.

// We therefore aggregate all successful lookups and treat

// per-object errors as yielding no results.

let mut all_objs: Vec<DMLObject> = Vec::new();

for res in curr_objs

.into_iter()

.map(|o| self.context_to_objs(o, context_chain, limitations))

{

match res {

Ok(mut objs) => all_objs.append(&mut objs),

Err(_) => {

// Swallow expected-miss errors; they do not cause

// the overall lookup to fail.

}

}

}

Ok(all_objs)

Copilot · 2026-04-01T14:29:00Z

USAGE.md

 `goto-implementations`)

+## Relevant environment variables
+* `MAX_LOG_MESSAGE_LENGTH` When set, will truncate any outputted logs that are longer than the value. Defaults to 1000 bytes. Set to 0 to turn off truncation.


The docs say logs will be truncated to MAX_LOG_MESSAGE_LENGTH, but the implementation currently slices to max_len and then appends " ...", producing output longer than the configured value when truncation occurs. Please clarify the behavior here (either adjust docs to mention the suffix, or adjust truncation logic to keep the final output within the configured limit).

Suggested change

* `MAX_LOG_MESSAGE_LENGTH` When set, will truncate any outputted logs that are longer than the value. Defaults to 1000 bytes. Set to 0 to turn off truncation.

* `MAX_LOG_MESSAGE_LENGTH` When set, will truncate any outputted log messages that are longer than the value and append `" ..."`. Up to `MAX_LOG_MESSAGE_LENGTH` bytes of the original message are preserved; the suffix may cause the final output to slightly exceed this value. Defaults to 1000 bytes. Set to 0 to turn off truncation.

JonatanWaern force-pushed the rework-early-cancellation branch from 74bb34b to 8ac4862 Compare April 1, 2026 12:53

JonatanWaern mentioned this pull request Apr 1, 2026

Add env variable for truncating max log length #194

Closed

JonatanWaern added 2 commits April 1, 2026 14:55

Box large enum variants

19e5e05

Signed-off-by: Jonatan Waern <jonatan.waern@intel.com>

Refine obj-to-contexts mapping error propagation

f63bc2d

Silences internal errors from things that are not actually such, meaning we can return empty object results from contexts when we think it's fine Signed-off-by: Jonatan Waern <jonatan.waern@intel.com>

JonatanWaern force-pushed the rework-early-cancellation branch from 8ac4862 to 63c0242 Compare April 1, 2026 12:55

JonatanWaern marked this pull request as ready for review April 1, 2026 12:57

JonatanWaern requested review from TSonono and Copilot April 1, 2026 12:57

Copilot started reviewing on behalf of JonatanWaern April 1, 2026 12:58 View session

Copilot AI reviewed Apr 1, 2026

View reviewed changes

JonatanWaern force-pushed the rework-early-cancellation branch from 63c0242 to 5362d06 Compare April 1, 2026 14:19

JonatanWaern added 5 commits April 1, 2026 16:20

Add setting to cap reference cache size

f30874f

Signed-off-by: Jonatan Waern <jonatan.waern@intel.com>

Terminate jobs via result return, rather than panic

5d778fa

Signed-off-by: Jonatan Waern <jonatan.waern@intel.com>

Disable recursive cache

9aa8bc7

This is meant to disallow infinite recursion when something breaks, but the resource overhead has proven too large (15%~ of peak mem foorprint)

Terminate message reader faster

1a396a7

Signed-off-by: Jonatan Waern <jonatan.waern@intel.com>

Add env variable to truncate log length

5e0222e

Signed-off-by: Jonatan Waern <jonatan.waern@intel.com>

JonatanWaern force-pushed the rework-early-cancellation branch from 5362d06 to 5e0222e Compare April 1, 2026 14:22

JonatanWaern requested a review from Copilot April 1, 2026 14:22

Copilot started reviewing on behalf of JonatanWaern April 1, 2026 14:23 View session

Copilot AI reviewed Apr 1, 2026

View reviewed changes

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Optimization of Device Analysis#217

Optimization of Device Analysis#217
JonatanWaern wants to merge 7 commits intomainfrom
rework-early-cancellation

JonatanWaern commented Mar 25, 2026

Uh oh!

Copilot AI left a comment

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Copilot AI left a comment

Uh oh!

Copilot AI Apr 1, 2026

Uh oh!

Copilot AI Apr 1, 2026

Uh oh!

Copilot AI Apr 1, 2026

Uh oh!

Copilot AI Apr 1, 2026

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

	- Added environment variable to control max per-log output length for most logging from the DLS, defaulting to 1000 bytes..
	- Added environment variable to control max per-log output length for most logging from the DLS, defaulting to 1000 bytes.

-        // Note: failing to find here is not actually an error
-        curr_objs.into_iter()
-            .map(|o|self.context_to_objs(o, context_chain, limitations))
-            .process_results(|r|r.flatten().collect())
+        // Note: failing to find here is not actually an error.
+        // We therefore aggregate all successful lookups and treat
+        // per-object errors as yielding no results.
+        let mut all_objs: Vec<DMLObject> = Vec::new();
+        for res in curr_objs
+            .into_iter()
+            .map(|o| self.context_to_objs(o, context_chain, limitations))
+        {
+            match res {
+                Ok(mut objs) => all_objs.append(&mut objs),
+                Err(_) => {
+                    // Swallow expected-miss errors; they do not cause
+                    // the overall lookup to fail.
+                }
+            }
+        }
+        Ok(all_objs)

	* `MAX_LOG_MESSAGE_LENGTH` When set, will truncate any outputted logs that are longer than the value. Defaults to 1000 bytes. Set to 0 to turn off truncation.
	* `MAX_LOG_MESSAGE_LENGTH` When set, will truncate any outputted log messages that are longer than the value and append `" ..."`. Up to `MAX_LOG_MESSAGE_LENGTH` bytes of the original message are preserved; the suffix may cause the final output to slightly exceed this value. Defaults to 1000 bytes. Set to 0 to turn off truncation.

Conversation

JonatanWaern commented Mar 25, 2026

Uh oh!

Copilot AI left a comment

Choose a reason for hiding this comment

Pull request overview

Reviewed changes

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Copilot AI left a comment

Choose a reason for hiding this comment

Pull request overview

Uh oh!

Copilot AI Apr 1, 2026

Choose a reason for hiding this comment

Uh oh!

Copilot AI Apr 1, 2026

Choose a reason for hiding this comment

Uh oh!

Copilot AI Apr 1, 2026

Choose a reason for hiding this comment

Uh oh!

Copilot AI Apr 1, 2026

Choose a reason for hiding this comment

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants