Excel add-in: Bloomberg-grade overhaul of the bond pricing surface - #127
Conversation
Correctness and safety: - Catch panics at every FFI export and ship cdylibs with the new [profile.excel] (panic=unwind), so an engine panic surfaces as an in-cell error instead of killing the host; Excel and Java builds both use the profile - Native Excel errors (#VALUE!/#REF!/#NUM!/#NAME?/#N/A) with per-cell detail via CX.LASTERROR, CX.DIAG, and a ribbon error log - Strict culture-invariant parsing: embedded blanks or unparseable cells in parallel ranges error with their position instead of silently misaligning; dates never parse under the machine locale Performance (targets 1k-10k pricing cells per workbook): - Idempotent registration: rebuilding an unchanged spec returns the same handle with no generation bump, so Excel dependency cutoff stops recalc cascades; edited specs still evict and re-mint - Generation-gated response cache keyed (verb, request): multi-field blocks cost one engine call, repeated recalcs are near-free - One JSON parse per RPC; pointer-scan UTF-8 marshalling Usability: - Ticker-style referencing: analytics UDFs accept CUSIP/ISIN/name or curve names via a registry alias table; auto rebuild on add-in load heals reopened workbooks (ribbon Rebuild for manual use) - CX.YAS one-call yield & spread analysis grid; CX.SCENARIO curve ladder in one call, zero-shift row anchored at the mark, with a curveless yield-shift fallback in the Scenario form - CX.PRICE/RISK/SPREAD/YAS .LIVE variants: RTD-backed cells that recompute automatically when referenced objects change - IntelliSense tooltips packed into the .xll; modeless single-instance tickets; hedge advisor exposed (CX.RISKPROFILE, CX.HEDGE) Engine: Z-spread solves now round to 1e-4 bp instead of whole bp, so repricing at the returned spread reproduces the input price to well under a cent. Tests: headless suites on both sides (42 Rust incl. registry idempotency, panic guards, ticker resolution, YAS/scenario smoke; 46 C# covering parsing, culture invariance, cache generation gating, error mapping). Manual Excel checklist extended in SMOKE_TEST.md.
|
Warning Review limit reached
Next review available in: 31 minutes Enable usage-based reviews in Billing to review now. Otherwise, wait until the next included review is available. How can I continue?After more reviews become available, a review can be triggered using the To avoid repeated limits, reduce automatic review volume by pausing incremental auto-reviews earlier, using label-based review opt-in, excluding WIP or generated PR titles, or requesting reviews manually when the PR is ready. If your team needs uninterrupted high-volume reviews, an organization admin can enable usage-based reviews. How do review limits work?CodeRabbit enforces per-developer PR review limits for each organization. Most developers receive the normal plan review availability. For paid Pro and Pro+ PR reviews, CodeRabbit uses adaptive limits for sustained high-volume activity. When a developer's recent PR review activity reaches the 95th percentile or higher among CodeRabbit users, additional reviews become available more gradually as earlier reviews age out of the rolling window. Please refer docs for additional details. Review details⚙️ Run configurationConfiguration used: defaults Review profile: CHILL Plan: Pro Run ID: 📒 Files selected for processing (28)
📝 WalkthroughWalkthroughThis change adds an Excel-specific unwinding Cargo profile, alias-aware idempotent FFI registration, new YAS and scenario analytics, panic-safe native boundaries, generation-aware Excel caching, live UDFs, diagnostics, expanded forms, and corresponding tests and documentation. ChangesExcel and FFI integration
Estimated code review effort: 5 (Critical) | ~120 minutes Possibly related PRs
🚥 Pre-merge checks | ✅ 4 | ❌ 1❌ Failed checks (1 warning)
✅ Passed checks (4 passed)
✨ Finishing Touches🧪 Generate unit tests (beta)
Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out. Comment |
Cut narration, design essays, and restated rationale across the add-in and FFI crates; comments now carry only what the code cannot show (threading rules, panic-profile requirement, locale traps, race windows, eviction semantics). Also refreshed the stale UDF-surface banner in Functions.cs.
![cubic-dev-ai[bot]](https://avatars.githubusercontent.com/in/1082092?s=60&v=4){kind=small}
There was a problem hiding this comment.
2 issues found and verified against the latest diff
Prompt for AI agents (unresolved issues)
Check if these issues are valid — if so, understand the root cause of each and fix them. If appropriate, use sub-agents to investigate and fix each issue separately.
<file name="crates/convex-analytics/src/dto.rs">
<violation number="1" location="crates/convex-analytics/src/dto.rs:623">
P2: Schema consumers cannot discover or validate `convex_scenario`: `convex_schema("ScenarioRequest")` fails because these DTOs were not added to the curated schema registry. Add request/response schema definitions and lookup arms alongside this public wire contract.</violation>
</file>
<file name="excel/README.md">
<violation number="1" location="excel/README.md:299">
P2: The README's error-mapping table documents `#REF!` for unknown handles/tickers, but the SMOKE_TEST expects `#VALUE!` for unknown handle strings until ticker lookup is fully functional. If a user tries `=CX.PRICE("BAD_HANDLE", …)` and gets `#VALUE!`, they'll find the README says unknown handles yield `#REF!` — there's no note explaining the transitional state. Either update the README to reflect the current error mapping (and note that ticker-resolution will switch the error code), or add a transitional footnote so users aren't misled.</violation>
</file>
Tip: cubic can generate docs of your entire codebase and keep them up to date. Try it here.
Re-trigger cubic
| |---|---| | ||
| | `#VALUE!` | malformed input — bad mark, unparseable cell, misaligned parallel ranges | | ||
| | `#NAME?` | unknown keyword — frequency, day count, spread type, field/metric name | | ||
| | `#REF!` | handle (or ticker) not registered, or wrong object kind | |
There was a problem hiding this comment.
P2: The README's error-mapping table documents #REF! for unknown handles/tickers, but the SMOKE_TEST expects #VALUE! for unknown handle strings until ticker lookup is fully functional. If a user tries =CX.PRICE("BAD_HANDLE", …) and gets #VALUE!, they'll find the README says unknown handles yield #REF! — there's no note explaining the transitional state. Either update the README to reflect the current error mapping (and note that ticker-resolution will switch the error code), or add a transitional footnote so users aren't misled.
Prompt for AI agents
Check if this issue is valid — if so, understand the root cause and fix it. At excel/README.md, line 299:
<comment>The README's error-mapping table documents `#REF!` for unknown handles/tickers, but the SMOKE_TEST expects `#VALUE!` for unknown handle strings until ticker lookup is fully functional. If a user tries `=CX.PRICE("BAD_HANDLE", …)` and gets `#VALUE!`, they'll find the README says unknown handles yield `#REF!` — there's no note explaining the transitional state. Either update the README to reflect the current error mapping (and note that ticker-resolution will switch the error code), or add a transitional footnote so users aren't misled.</comment>
<file context>
@@ -237,9 +286,23 @@ Codes:
+|---|---|
+| `#VALUE!` | malformed input — bad mark, unparseable cell, misaligned parallel ranges |
+| `#NAME?` | unknown keyword — frequency, day count, spread type, field/metric name |
+| `#REF!` | handle (or ticker) not registered, or wrong object kind |
+| `#NUM!` | solver non-convergence or domain error |
+| `#N/A` | native library not loaded — run `=CX.DIAG()` |
</file context>
![coderabbitai[bot]](https://avatars.githubusercontent.com/in/347564?s=60&v=4){kind=small}
There was a problem hiding this comment.
Actionable comments posted: 4
🧹 Nitpick comments (2)
excel/Convex.Excel/forms/PricingTicketForm.cs (1)
155-169: 📐 Maintainability & Code Quality | 🔵 Trivial | ⚡ Quick win
BuildRequestandBuildFormula/StampRefindependently re-derive the same fields.Both methods separately pull bond/curve/settle/mark/frequency from the UI. If a new field is ever added to one and not the other, the ticket could validate one request while stamping a formula for a different one.
Consider extracting a single "gather ticket inputs" step (bond/curve refs, settle date, mark, frequency) that both the JSON request builder and the formula builder consume, so the two representations can't drift.
Also applies to: 189-220
🤖 Prompt for AI Agents
Verify each finding against current code. Fix only still-valid issues, skip the rest with a brief reason, keep changes minimal, and validate. In `@excel/Convex.Excel/forms/PricingTicketForm.cs` around lines 155 - 169, The ticket inputs are independently gathered by BuildRequest and BuildFormula/StampRef, allowing their representations to diverge. Extract a shared input-gathering method or model for the bond and optional curve references, settlement date, mark, and quote frequency, then have BuildRequest and the formula/stamping flow consume that single result while preserving the existing validation and optional-curve behavior.crates/convex-ffi/src/dispatch.rs (1)
116-131: 📐 Maintainability & Code Quality | 🔵 Trivial | ⚡ Quick winAdd a guard test for resolver tables
These
*_REFSlists are still hand-maintained, so a newHandle/Option<Handle>field can slip past alias resolution and degrade to a genericserde_json::from_valueerror. A small test that asserts each request DTO’s handle-bearing fields are covered would make drift much harder;SPREAD_REFS’ nestedparams.*paths make this especially easy to miss.🤖 Prompt for AI Agents
Verify each finding against current code. Fix only still-valid issues, skip the rest with a brief reason, keep changes minimal, and validate. In `@crates/convex-ffi/src/dispatch.rs` around lines 116 - 131, Add a guard test alongside the resolver-table definitions that inspects every request DTO’s handle-bearing fields and asserts they are represented in the corresponding *_REFS table, including nested params paths such as SPREAD_REFS. Cover all listed request types and both Handle and Option<Handle> fields so future DTO changes fail the test until alias resolution is updated.
🤖 Prompt for all review comments with AI agents
Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.
Inline comments:
In `@crates/convex-ffi/src/build.rs`:
- Around line 427-438: Update curve_name_and_aliases so registry_key and name
are trimmed and treated as absent when empty before selecting reg_name. Preserve
the existing alias filtering, and ensure blank inputs cannot produce an empty
registry name or cause unrelated curves to share the same eviction slot.
In `@crates/convex-ffi/src/registry.rs`:
- Around line 91-98: Update Inner.tables and all registry accessors used by
register(), release(), clear_all(), and read paths to avoid permanent RwLock
poisoning: either replace it with a non-poisoning lock or recover poisoned
read/write guards explicitly. Ensure panic guards still return errors while
subsequent registry operations can continue after a panic.
In `@excel/Convex.Excel/CxCache.cs`:
- Line 43: Replace the concatenated string key in the cache-key construction
with a collision-safe tuple containing both verb and requestJson. Update the
surrounding cache access in CxCache so the tuple is used consistently for lookup
and storage.
In `@excel/README.md`:
- Line 103: Update the CX.SPREAD documentation and its associated settings
dropdown so the accepted ASW/ASW_PAR aliases and CREDIT option are represented
consistently; either document both aliases and CREDIT clearly or add the missing
selectable options, while preserving the existing default field behavior.
---
Nitpick comments:
In `@crates/convex-ffi/src/dispatch.rs`:
- Around line 116-131: Add a guard test alongside the resolver-table definitions
that inspects every request DTO’s handle-bearing fields and asserts they are
represented in the corresponding *_REFS table, including nested params paths
such as SPREAD_REFS. Cover all listed request types and both Handle and
Option<Handle> fields so future DTO changes fail the test until alias resolution
is updated.
In `@excel/Convex.Excel/forms/PricingTicketForm.cs`:
- Around line 155-169: The ticket inputs are independently gathered by
BuildRequest and BuildFormula/StampRef, allowing their representations to
diverge. Extract a shared input-gathering method or model for the bond and
optional curve references, settlement date, mark, and quote frequency, then have
BuildRequest and the formula/stamping flow consume that single result while
preserving the existing validation and optional-curve behavior.
🪄 Autofix (Beta)
Fix all unresolved CodeRabbit comments on this PR:
- Push a commit to this branch (recommended)
- Create a new PR with the fixes
ℹ️ Review info
⚙️ Run configuration
Configuration used: defaults
Review profile: CHILL
Plan: Pro
Run ID: 517c0077-126a-42ef-ad73-d49935a8ca2f
📒 Files selected for processing (44)
.github/workflows/ci.yml.github/workflows/java-release.yml.github/workflows/java.yml.github/workflows/release.yml.gitignoreCargo.tomlcrates/convex-analytics/src/dto.rscrates/convex-analytics/src/spreads/zspread.rscrates/convex-ffi/src/build.rscrates/convex-ffi/src/dispatch.rscrates/convex-ffi/src/lib.rscrates/convex-ffi/src/registry.rscrates/convex-ffi/src/schemas.rscrates/convex-ffi/tests/smoke.rsexcel/Convex.Excel.Tests/Convex.Excel.Tests.csprojexcel/Convex.Excel.Tests/CxCacheTests.csexcel/Convex.Excel.Tests/CxParseTests.csexcel/Convex.Excel/Convex.Excel.csprojexcel/Convex.Excel/Convex.Excel.dnaexcel/Convex.Excel/Cx.csexcel/Convex.Excel/CxCache.csexcel/Convex.Excel/CxCaller.csexcel/Convex.Excel/CxErrorStore.csexcel/Convex.Excel/CxLive.csexcel/Convex.Excel/CxParse.csexcel/Convex.Excel/CxSettings.csexcel/Convex.Excel/Functions.csexcel/Convex.Excel/NativeLoader.csexcel/Convex.Excel/RibbonController.csexcel/Convex.Excel/forms/BondBuilderForm.csexcel/Convex.Excel/forms/CurveBuilderForm.csexcel/Convex.Excel/forms/CurveViewerForm.csexcel/Convex.Excel/forms/ErrorLogForm.csexcel/Convex.Excel/forms/ObjectBrowserForm.csexcel/Convex.Excel/forms/PricingTicketForm.csexcel/Convex.Excel/forms/ScenarioForm.csexcel/Convex.Excel/forms/SchemaBrowserForm.csexcel/Convex.Excel/forms/SettingsForm.csexcel/Convex.Excel/forms/SpreadTicketForm.csexcel/Convex.Excel/helpers/ComboReload.csexcel/Convex.Excel/helpers/FormUi.csexcel/Convex.Excel/helpers/SheetHelpers.csexcel/README.mdexcel/SMOKE_TEST.md
| fn curve_name_and_aliases( | ||
| registry_key: Option<String>, | ||
| name: Option<String>, | ||
| ) -> (Option<String>, Vec<String>) { | ||
| let aliases: Vec<String> = name | ||
| .iter() | ||
| .map(|s| s.trim().to_string()) | ||
| .filter(|s| !s.is_empty()) | ||
| .collect(); | ||
| let reg_name = registry_key.or(name); | ||
| (reg_name, aliases) | ||
| } |
There was a problem hiding this comment.
🗄️ Data Integrity & Integration | 🟡 Minor | ⚡ Quick win
curve_name_and_aliases doesn't filter empty-string keys before using them as the eviction slot.
aliases is trimmed and empty-filtered, but reg_name = registry_key.or(name) isn't — if either arrives as Some("") (e.g. a blank Excel argument marshaled as an empty string rather than omitted/null), it's used as-is as the registry name. Two unrelated curves both landing on registry_key: Some("") would collide in t.names, and register()'s idempotent-eviction logic would evict one when the other rebuilds, since it matches on the exact name string.
🐛 Proposed fix
fn curve_name_and_aliases(
registry_key: Option<String>,
name: Option<String>,
) -> (Option<String>, Vec<String>) {
- let aliases: Vec<String> = name
- .iter()
- .map(|s| s.trim().to_string())
- .filter(|s| !s.is_empty())
- .collect();
- let reg_name = registry_key.or(name);
+ let registry_key = registry_key.map(|s| s.trim().to_string()).filter(|s| !s.is_empty());
+ let name = name.map(|s| s.trim().to_string()).filter(|s| !s.is_empty());
+ let aliases: Vec<String> = name.clone().into_iter().collect();
+ let reg_name = registry_key.or(name);
(reg_name, aliases)
}📝 Committable suggestion
‼️ IMPORTANT
Carefully review the code before committing. Ensure that it accurately replaces the highlighted code, contains no missing lines, and has no issues with indentation. Thoroughly test & benchmark the code to ensure it meets the requirements.
| fn curve_name_and_aliases( | |
| registry_key: Option<String>, | |
| name: Option<String>, | |
| ) -> (Option<String>, Vec<String>) { | |
| let aliases: Vec<String> = name | |
| .iter() | |
| .map(|s| s.trim().to_string()) | |
| .filter(|s| !s.is_empty()) | |
| .collect(); | |
| let reg_name = registry_key.or(name); | |
| (reg_name, aliases) | |
| } | |
| fn curve_name_and_aliases( | |
| registry_key: Option<String>, | |
| name: Option<String>, | |
| ) -> (Option<String>, Vec<String>) { | |
| let registry_key = registry_key | |
| .map(|s| s.trim().to_string()) | |
| .filter(|s| !s.is_empty()); | |
| let name = name | |
| .map(|s| s.trim().to_string()) | |
| .filter(|s| !s.is_empty()); | |
| let aliases: Vec<String> = name.clone().into_iter().collect(); | |
| let reg_name = registry_key.or(name); | |
| (reg_name, aliases) | |
| } |
🤖 Prompt for AI Agents
Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.
In `@crates/convex-ffi/src/build.rs` around lines 427 - 438, Update
curve_name_and_aliases so registry_key and name are trimmed and treated as
absent when empty before selecting reg_name. Preserve the existing alias
filtering, and ensure blank inputs cannot produce an empty registry name or
cause unrelated curves to share the same eviction slot.
| struct Inner { | ||
| next_handle: AtomicU64, | ||
| /// Bumped on every mutation (register of new content, release, clear) and | ||
| /// NOT on idempotent re-registration. Callers key caches on this: same | ||
| /// generation ⇒ every handle resolves to the same object it did before. | ||
| generation: AtomicU64, | ||
| tables: RwLock<Tables>, | ||
| } |
There was a problem hiding this comment.
🩺 Stability & Availability | 🟠 Major | ⚡ Quick win
🧩 Analysis chain
🏁 Script executed:
#!/bin/bash
set -euo pipefail
# Inspect the registry implementation and panic guards.
git ls-files crates/convex-ffi/src/registry.rs crates/convex-ffi/src/lib.rs
printf '\n--- registry.rs (outline) ---\n'
ast-grep outline crates/convex-ffi/src/registry.rs --view expanded || true
printf '\n--- lib.rs (outline) ---\n'
ast-grep outline crates/convex-ffi/src/lib.rs --view expanded || true
printf '\n--- registry.rs relevant sections ---\n'
sed -n '1,220p' crates/convex-ffi/src/registry.rs
printf '\n--- registry.rs later sections ---\n'
sed -n '220,420p' crates/convex-ffi/src/registry.rs
printf '\n--- lib.rs relevant sections ---\n'
sed -n '1,240p' crates/convex-ffi/src/lib.rsRepository: sujitn/convex
Length of output: 26300
Avoid lock poisoning in REGISTRY.tables
register(), release(), and clear_all() can poison the RwLock on any panic, and the later .read().unwrap() / .write().unwrap() calls will keep failing for the rest of the process. The panic guards only convert the first failure into an error; they do not recover the poisoned lock. Switch to a non-poisoning lock or recover the poison explicitly.
🤖 Prompt for AI Agents
Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.
In `@crates/convex-ffi/src/registry.rs` around lines 91 - 98, Update Inner.tables
and all registry accessors used by register(), release(), clear_all(), and read
paths to avoid permanent RwLock poisoning: either replace it with a
non-poisoning lock or recover poisoned read/write guards explicitly. Ensure
panic guards still return errors while subsequent registry operations can
continue after a panic.
| } | ||
| } | ||
|
|
||
| var key = verb + "" + requestJson; |
There was a problem hiding this comment.
🗄️ Data Integrity & Integration | 🟡 Minor | ⚡ Quick win
Cache key has no real delimiter between verb and requestJson.
verb + "" + requestJson is equivalent to verb + requestJson — the empty string contributes nothing. This risks key collisions between different (verb, requestJson) pairs that happen to concatenate to the same string (e.g. if a future verb name is a prefix/suffix overlap of another, or a request body isn't a JSON object). In a response cache for pricing/spread results, a collision means silently returning a cached result for the wrong request.
🔧 Proposed fix: use a tuple key instead of string concatenation
- private static readonly ConcurrentDictionary<string, string> _cache = new();
+ private static readonly ConcurrentDictionary<(string Verb, string RequestJson), string> _cache = new();
@@
- var key = verb + "" + requestJson;
+ var key = (verb, requestJson);🤖 Prompt for AI Agents
Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.
In `@excel/Convex.Excel/CxCache.cs` at line 43, Replace the concatenated string
key in the cache-key construction with a collision-safe tuple containing both
verb and requestJson. Update the surrounding cache access in CxCache so the
tuple is used consistently for lookup and storage.
| | `CX.PRICE(bond, settle, mark, [curve], [quoteFreq], [field])` | Returns clean / dirty / accrued / YTM / Z-bps. `field`: `clean` (default), `dirty`, `accrued`, `ytm`, `z_spread`, `grid`. | | ||
| | `CX.YAS(bond, settle, mark, curve, [govtCurve], [quoteFreq], [swapCurve])` | **One-call Bloomberg-YAS-style grid**: every yield convention, G/Z/benchmark/ASW spreads, durations, convexity, DV01, and the settlement invoice — a single engine computation. | | ||
| | `CX.RISK(bond, settle, mark, [curve], [metric], [quoteFreq], [krdTenors])` | `metric`: `grid` (default), `mod_dur`, `mac_dur`, `convexity`, `dv01`, `spread_dur`, `krd`. | | ||
| | `CX.SPREAD(bond, curve, settle, mark, [type], [vol], [field])` | Z / G / I / ASW / ASW_PROC / OAS / DM / Credit. `field`: `bps` (default), `grid`. | |
There was a problem hiding this comment.
📐 Maintainability & Code Quality | 🟡 Minor | ⚡ Quick win
🧩 Analysis chain
🏁 Script executed:
#!/bin/bash
# Verify the actual spread type tokens accepted by the UDF and the ErrorCodes definition.
rg -rn "ASW_PAR|ASW\b|\"Credit\"" --type=cs -g '!**/bin/**' -g '!**/obj/**' -C2
# Also check the Rust DTO for spread type enum variants.
rg -rn "Asw|AswPar|Credit" --type=rust -C2Repository: sujitn/convex
Length of output: 151
🏁 Script executed:
#!/bin/bash
set -euo pipefail
# Locate the relevant files and map their contents lightly.
git ls-files | rg '(^|/)(SettingsForm\.cs|README\.md|.*spread.*|.*CX.*|.*UDF.*|.*Settings.*)$'
# Find every occurrence of the spread token names and dropdown labels.
rg -n --hidden --glob '!**/bin/**' --glob '!**/obj/**' --glob '!**/.git/**' \
'ASW_PAR|ASW\b|ASW_PROC|Credit|CX\.SPREAD|spread type|spreadType|SpreadType' .
# If SettingsForm.cs exists, print the surrounding lines around the reported label mapping.
if git ls-files | rg -q 'SettingsForm\.cs$'; then
file=$(git ls-files | rg 'SettingsForm\.cs$' | head -n 1)
echo "== $file =="
rg -n -C3 'ASW_PAR|ASW\b|Credit|spread' "$file"
fi
# Show the README snippet around the documented CX.SPREAD line.
if git ls-files | rg -q 'excel/README\.md$'; then
file=excel/README.md
echo "== $file =="
rg -n -C2 'CX\.SPREAD' "$file"
fiRepository: sujitn/convex
Length of output: 50369
🏁 Script executed:
#!/bin/bash
set -euo pipefail
echo "== excel/Convex.Excel/CxParse.cs =="
rg -n -C4 'AsSpreadType|ASW_PAR|CREDIT|AssetSwapPar|AssetSwapProceeds' excel/Convex.Excel/CxParse.cs
echo
echo "== excel/Convex.Excel/forms/SettingsForm.cs =="
rg -n -C4 'Items\.AddRange|Default spread|DefaultSpreadType|ASW_PAR|ASW_PROC|Credit' excel/Convex.Excel/forms/SettingsForm.cs
echo
echo "== crates/convex-core/src/types/mark.rs =="
rg -n -C4 'parse_spread_type|ASW_PAR|CREDIT|AssetSwapPar|AssetSwapProceeds' crates/convex-core/src/types/mark.rs
echo
echo "== excel/README.md context =="
rg -n -C2 'Spread types:|CX\.SPREAD\(' excel/README.mdRepository: sujitn/convex
Length of output: 8004
Clarify the ASW alias and missing default option
CX.SPREAD accepts both ASW and ASW_PAR, but the settings dropdown only exposes ASW_PAR, and CREDIT is accepted by the parser yet not selectable there. Either document the aliases or add the missing UI option so the terminology stays consistent.
🤖 Prompt for AI Agents
Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.
In `@excel/README.md` at line 103, Update the CX.SPREAD documentation and its
associated settings dropdown so the accepted ASW/ASW_PAR aliases and CREDIT
option are represented consistently; either document both aliases and CREDIT
clearly or add the missing selectable options, while preserving the existing
default field behavior.
- YAS: G-/I-spread marks now price against the supplied government/swap
curve instead of always the spot curve (with an end-to-end test)
- All-digit tickers ("037833100") resolve as names; bare digit strings
are no longer misparsed as numeric handles
- CxCache: entries are generation-tagged and validated on every hit
(closes the stale-hit race), and Lazy values give single-flight so
concurrent identical calls share one engine call
- Live cells record errors under their owning cell (captured at
subscribe time), so CX.LASTERROR works after timer-driven refreshes
- Curve viewer: renders invalidate outstanding sweeps, obsolete sweeps
stop mid-loop, and the close/BeginInvoke race is guarded
- Error log jump handles sheet names containing apostrophes; combo
labels keep names containing the display separator intact
- Scenario form gains key_rate with a pivot/key-tenor field
- Pre-dynamic-array Excel: stamped grids fall back to a legacy CSE
array formula instead of showing one cell
- ScenarioRequest/Response added to the schema registry; Z-spread
Decimals built with from_f64 (clean digits); docs aligned (#REF! for
unknown tickers, panic → #VALUE!)
There was a problem hiding this comment.
4 issues found across 21 files (changes from recent commits).
Prompt for AI agents (unresolved issues)
Check if these issues are valid — if so, understand the root cause of each and fix them. If appropriate, use sub-agents to investigate and fix each issue separately.
<file name="crates/convex-ffi/src/schemas.rs">
<violation number="1" location="crates/convex-ffi/src/schemas.rs:297">
P2: Schema-based clients can validate a scenario bump then receive `invalid_input` from `convex_scenario`, because required fields are conditional on `kind` but the schema requires only `kind`. Represent each bump variant as a conditional/`oneOf` schema with its required inputs.</violation>
</file>
<file name="excel/Convex.Excel/CxLive.cs">
<violation number="1" location="excel/Convex.Excel/CxLive.cs:25">
P2: When identical `.LIVE` formulas appear in multiple cells, an RPC or RTD-push error is attached only to the cell that created the shared topic, so `CX.LASTERROR` for the other error cells is empty or misleading. Keep per-subscriber cell context for error recording, or use a cell-specific observation subscription while preserving the response cache.</violation>
</file>
<file name="excel/Convex.Excel/helpers/SheetHelpers.cs">
<violation number="1" location="excel/Convex.Excel/helpers/SheetHelpers.cs:51">
P2: Legacy Excel stamping fails for valid long ticker/name inputs because `Range.FormulaArray` accepts at most 255 characters. Validate this before the fallback and provide a supported legacy path or a clear, input-specific error rather than attempting the assignment.</violation>
<violation number="2" location="excel/Convex.Excel/helpers/SheetHelpers.cs:51">
P1: On pre-dynamic-array Excel, stamping into a cell with data in any of the next four rows or second column overwrites that data with the CSE formula. Check the target 5×2 range is empty (or require explicit confirmation) before assigning `FormulaArray`, matching the non-destructive `#SPILL!` behavior on newer Excel.</violation>
</file>
Tip: Review your code locally with the cubic CLI to iterate faster.
Re-trigger cubic
| catch | ||
| { | ||
| dynamic range = anchor.Resize(spillRows, spillCols); | ||
| range.FormulaArray = formula; |
There was a problem hiding this comment.
P1: On pre-dynamic-array Excel, stamping into a cell with data in any of the next four rows or second column overwrites that data with the CSE formula. Check the target 5×2 range is empty (or require explicit confirmation) before assigning FormulaArray, matching the non-destructive #SPILL! behavior on newer Excel.
Prompt for AI agents
Check if this issue is valid — if so, understand the root cause and fix it. At excel/Convex.Excel/helpers/SheetHelpers.cs, line 51:
<comment>On pre-dynamic-array Excel, stamping into a cell with data in any of the next four rows or second column overwrites that data with the CSE formula. Check the target 5×2 range is empty (or require explicit confirmation) before assigning `FormulaArray`, matching the non-destructive `#SPILL!` behavior on newer Excel.</comment>
<file context>
@@ -33,6 +33,26 @@ public static string WriteFormulaAtSelection(string formula)
+ catch
+ {
+ dynamic range = anchor.Resize(spillRows, spillCols);
+ range.FormulaArray = formula;
+ }
+ return anchor.Address;
</file context>
| "name": {"type": ["string","null"]}, | ||
| "bumps": {"type": "array", "items": { | ||
| "type": "object", | ||
| "required": ["kind"], |
There was a problem hiding this comment.
P2: Schema-based clients can validate a scenario bump then receive invalid_input from convex_scenario, because required fields are conditional on kind but the schema requires only kind. Represent each bump variant as a conditional/oneOf schema with its required inputs.
Prompt for AI agents
Check if this issue is valid — if so, understand the root cause and fix it. At crates/convex-ffi/src/schemas.rs, line 297:
<comment>Schema-based clients can validate a scenario bump then receive `invalid_input` from `convex_scenario`, because required fields are conditional on `kind` but the schema requires only `kind`. Represent each bump variant as a conditional/`oneOf` schema with its required inputs.</comment>
<file context>
@@ -275,6 +277,61 @@ const YAS_RESPONSE: &str = r##"{
+ "name": {"type": ["string","null"]},
+ "bumps": {"type": "array", "items": {
+ "type": "object",
+ "required": ["kind"],
+ "properties": {
+ "kind": {"enum": ["parallel","steepener","flattener","key_rate","credit_spread"]},
</file context>
| var key = caller == null ? null : CxCaller.StableKey(caller); | ||
| var address = caller == null ? null : CxCaller.Describe(caller); | ||
| return ExcelAsyncUtil.Observe("CX.LIVE:" + verb, requestJson, | ||
| () => new RpcObservable(verb, requestJson, select, key, address)); |
There was a problem hiding this comment.
P2: When identical .LIVE formulas appear in multiple cells, an RPC or RTD-push error is attached only to the cell that created the shared topic, so CX.LASTERROR for the other error cells is empty or misleading. Keep per-subscriber cell context for error recording, or use a cell-specific observation subscription while preserving the response cache.
Prompt for AI agents
Check if this issue is valid — if so, understand the root cause and fix it. At excel/Convex.Excel/CxLive.cs, line 25:
<comment>When identical `.LIVE` formulas appear in multiple cells, an RPC or RTD-push error is attached only to the cell that created the shared topic, so `CX.LASTERROR` for the other error cells is empty or misleading. Keep per-subscriber cell context for error recording, or use a cell-specific observation subscription while preserving the response cache.</comment>
<file context>
@@ -16,34 +16,44 @@ internal static class CxLive
- CxErrorStore.Record(ex);
- return ErrorMapper.ToExcelError(ex);
- }
+ () => new RpcObservable(verb, requestJson, select, key, address));
}
</file context>
| catch | ||
| { | ||
| dynamic range = anchor.Resize(spillRows, spillCols); | ||
| range.FormulaArray = formula; |
There was a problem hiding this comment.
P2: Legacy Excel stamping fails for valid long ticker/name inputs because Range.FormulaArray accepts at most 255 characters. Validate this before the fallback and provide a supported legacy path or a clear, input-specific error rather than attempting the assignment.
Prompt for AI agents
Check if this issue is valid — if so, understand the root cause and fix it. At excel/Convex.Excel/helpers/SheetHelpers.cs, line 51:
<comment>Legacy Excel stamping fails for valid long ticker/name inputs because `Range.FormulaArray` accepts at most 255 characters. Validate this before the fallback and provide a supported legacy path or a clear, input-specific error rather than attempting the assignment.</comment>
<file context>
@@ -33,6 +33,26 @@ public static string WriteFormulaAtSelection(string formula)
+ catch
+ {
+ dynamic range = anchor.Resize(spillRows, spillCols);
+ range.FormulaArray = formula;
+ }
+ return anchor.Address;
</file context>
1 participant
Reworks the Excel add-in around how a terminal user actually works: reference bonds by ticker, get full analysis grids from one call, see live updates, and never lose the session to an engine fault.
Safety & correctness
catch_unwind; shipping cdylibs build with the new[profile.excel](panic=unwind), so a stray engine panic becomes an in-cell error instead of terminating Excel. The Java workflows build with the same profile for the same reason.#VALUE!/#REF!/#NUM!/#NAME?/#N/A) soIFERROR/ISERRORwork; full detail per cell via=CX.LASTERROR(cell),=CX.DIAG(), and a ribbon Error Log with jump-to-cell.Performance
convex_generation) gates a client-side response cache keyed(verb, request): a five-field price block costs one engine call, and repeated recalcs are near-free.Usability
=CX.PRICE("912828YK0", …)), resolved against a registry alias table; unknown names return#REF!. Auto rebuild on add-in load heals reopened workbooks; ribbon Rebuild covers workbooks opened mid-session.CX.YAS— one-call yield & spread analysis grid (street/current/simple/MM yields, G/Z/benchmark/ASW spreads, durations, convexity, DV01, settlement invoice).CX.SCENARIO— a whole shift ladder (parallel/steepener/flattener/key-rate/credit) in one call, zero-shift row anchored exactly at the mark; the Scenario form keeps a curveless yield-shift fallback..LIVEvariants of PRICE/RISK/SPREAD/YAS subscribe via RTD and recompute automatically when any referenced object is rebuilt or released.CX.RISKPROFILE,CX.HEDGE).Engine
ZSpreadCalculatornow returns spreads rounded to 1e-4 bp instead of whole basis points, so repricing at the solved spread reproduces the input price to well under a cent (previously a DV01-scale gap on long-duration bonds).Testing
cargo test -p convex-ffi) — registry idempotency/aliases/generation, panic guards, ticker resolution end-to-end, YAS and scenario smoke, JSON field-order hash stability.dotnet test excel/Convex.Excel.Tests) — strict parsing, culture invariance under a dd/MM locale, cache generation gating incl. a mutation-races-the-insert case, error-code mapping.excel/SMOKE_TEST.md(ticker refs, live cells, recalc cutoff, workbook reopen, modeless forms) — not yet run against a live Excel.Summary by CodeRabbit
New Features
Bug Fixes