Lime CRM Desktop Client 11.3.4502

Product

Lime CRM Desktop Client

Version

11.3.4502

Date published

2026-05-21

Platform

Windows

Availability

Manual installation, Reference

Cumulative release for 11.3.x covering everything since the 11.3.4483 baseline.

• Filter validation report. A new diagnostic action on a saved filter opens a side-by-side report showing the filter as written, the form the system rewrites it to (sub-filters resolved, conditions merged and optimised), and the record count returned at each stage - backend query before optimisation, backend query after optimisation, and a step-by-step client-side composition. Any discrepancy between what the user expected and what they got is traceable to a specific stage. Each top-level sub-filter carries its standalone match count, every field resolves to its human-readable name and option labels, and a structured diagnostic log captures the internal decisions made during sub-filter merging and SQL generation. Intended for the “this filter returned weirdly few/many records” support case.

• Filter validation report flags self-eliminating user logic. A user-authored filter whose conditions mathematically reduce to “matches everything” or “matches nothing” - typically two NOT IN conditions joined with OR, or two IN joined with AND over disjoint value lists; or any leaf arriving with an empty value list - now shows the affected leaves in the report's filter tree inside a dashed amber frame with an “(eliminated by optimiser)” badge whose tooltip names the rule that dropped it. For the operator-flipped pattern (same operator on both sides plus the opposite boolean) the diagnostic log adds an actionable hint such as “If the intent was to exclude both values, the surrounding group should be ALL OF (AND), not ANY OF (OR)”. The same collapse occurring inside a programmatically-merged filter (sub-filter fused into a host filter during evaluation) is logged at Info instead of Warning - the elimination there is an expected merge by-product rather than a usability problem.

• Inspectors with web-based action panes share the embedded browser engine across opens. The first inspector in a session sets the engine up; subsequent inspectors reuse it instead of building a new one each time. Open latency drops noticeably from the second inspector onwards, and the page's compiled JavaScript stays warm in the renderer's cache between opens so it doesn't have to be re-parsed every time.

• Inspector contents appear sooner after the page finishes loading. The hand-off between “navigation complete” and “show the page” was going through an extra internal queue hop that could add several hundred milliseconds when the application was busy. That hop is gone, so the inspector switches from blank to populated more promptly.

• VBA and binding work dispatches in priority order. Async work in the application now runs in priority buckets — click-driven work runs ahead of background data-source work, and idle-priority work only runs when the UI thread is otherwise quiet. Background items that have been waiting too long are promoted automatically so nothing is starved. Per-pump-cycle work is wall-clock-bounded so the UI thread stays responsive while the queue catches up. Solution developers can target a specific bucket by passing an optional priority argument to Application.RunEx — positive values mean foreground, 0 is normal, negative is background, and -10 or lower is idle. Existing scripts continue to default to normal.

• Filters that reference themselves no longer fail with internal errors. Filter expressions that loop back to themselves through a sub-filter chain — directly or transitively — are now detected and the self-referencing portion is skipped. The rest of the filter still runs. Previously these cycles produced either a generic internal error or a silently incorrect result.

• Some filter queries now use database indexes that they previously bypassed. Specific patterns in the generated SQL were wrapping indexed columns in functions that prevented SQL Server from doing an index seek. Three of those patterns — aggregate comparisons that already include zero, ORDER BY against non-nullable columns, and certain sub-filter null guards — now emit SQL that the optimizer can use indexes for. Expect noticeable speed-ups on the affected filter shapes.

• Switching between records no longer re-fetches the URL when the record has not actually changed. Inspector cards with a web action pane were re-running a backend URL lookup on every internal record-update notification, even when the assigned record was the same one already shown. The lookup now only runs when the record reference actually changes.

• QueryTimeout now bounds total wall-clock duration, not just the backend command time. Filter QueryTimeout previously governed only the time the backend had to answer; the client-side wait — initial grace period plus the simultaneous-queries queue — was outside the budget. For short timeouts behind a busy queue the perceived wait could be many times the configured value, even when the backend itself completed within budget. The budget is now resolved once up front and applied both to the queue wait and to the resulting backend command timeout. When the entire budget is consumed before the request leaves the client, the user sees an explicit “Wall-clock timeout expired while waiting in the client-side query queue” message rather than a generic failure — distinguishable in telemetry and support from a server-side timeout. Sub-filter timeouts no longer extend the top-level budget.

• Faster failure for misconfigured option fields. Auto-generated relation-option queries — synthesised when no hand-coded option query exists — now time out at 1 second instead of 10. A broken auto-generated query fails fast rather than freezing the inspector for the full 10-second window. Hand-coded option queries continue to use the 10-second default.

• Inspector controls and the embedded browser initialise in parallel. WebView2 environment startup, controller creation, profile setup, cookie sync, and the initial navigation now begin while the backend record fetch is still in flight, rather than being gated on the first paint of the host control. Inspectors with web-based action panes open with lower visible latency.

• Robustness improvements and better error messages around the embedded browser engine. When the engine fails to start within ten seconds — typically because another Lime instance is holding the user data folder, or because the WebView2 Runtime is corrupt or a renderer child is wedged — Lime now reports an actionable error naming the likely cause rather than letting the inspector hang indefinitely. Engine child processes (renderer, GPU, crashpad) are now closed in order during Lime's shutdown rather than left running as strays. The pool that shares the engine across inspectors also detects and recovers cleanly when WebView2 auto-updates itself between sessions.

• Action Pane renders as a unit. VBA bindings on the action pane are now drained in batches per UI pump iteration, and the slide-down animation that staggered each row's reveal has been removed. The pane appears assembled in one paint instead of visibly rendering itself row by row.

• DebugView trace for silently-dropped filter conditions. Filter conditions discarded during query generation — typically because the source field doesn't exist in the current schema — now emit a DebugView line identifying the discard, so configuration drift is easier to diagnose.

• Lime-link errors now identify which link failed. Errors raised while processing a limecrm:// link — out-of-memory conditions, internal-error catch-alls, and “no class specified” parse failures — now append the offending URL to the error message. The information also reaches the support telemetry channel as a structured event rather than a breadcrumb, so recurring failures in customer environments can be diagnosed without reproducing the case locally. The link URL is preserved across cross-process dispatch (when one Lime CRM instance hands a link off to another via the standard limecrm:// protocol handler), so the receiving instance can report the original URL even when the link arrived second-hand.

• Clearer dropdown status texts in link fields. The footer tooltip now says “loaded” rather than “matching” for the partial-results case — the larger number is the loaded set, the smaller is the matching subset, and “matching” was ambiguous about which it qualified. When a backend search returns no rows, the footer states “No matches” outright instead of pointing the user back at the keyboard shortcut they just pressed.

• Inspector splitter rework. The Fields tab now retains its state when you navigate between records, the inspector splitter supports smooth drag-resize with animated transitions, and the splitter position is remembered per inspector layout.

• Improved field layout in inspectors. Mixed rows that combine a multi-line field (such as a memo or expanded list) with single-line fields no longer leave awkward vertical gaps next to the single-line fields. Inspectors with an expandable field at the bottom now extend that field to fill the remaining space, matching typical form-layout expectations.

• Multi-line notes resize live with content. Multi-line text fields in the inspector — Notes (Anteckningar), History, Description, Memo — now grow and shrink as you add or remove content, without having to manually drag the splitter or scroll within the field. When the inspector is compressed, the field keeps at least three lines visible so small notes don't end up behind a scrollbar.

• Filter robustness for invalid option values. Pre-existing filters that reference option-set entries no longer present in the data model — values that have been renamed, removed, or otherwise can't be resolved — no longer break the query. Invalid entries are skipped, the surviving values continue to participate, and the filter still runs.

• Filter values resolve across locales and ID formats. Filter values authored against the canonical English label, or stored in the semicolon-wrapped ID form that some integrations produce, are now resolved automatically. The user's active locale no longer has to match the value source.

• Link-field dropdown search is faster and quieter. Typing into a link field's search box now debounces briefly before querying, so fast typing no longer fires one search per keystroke. Subsequent searches against the same field skip records already known from earlier matches, cutting redundant round-trips. The search box also auto-fills the longest common prefix of the matching options as a selected suggestion — keep typing to override, accept it as-is, or use it as a fast way to disambiguate similar entries.

• Long dropdown entries stay readable when the search needle falls in the middle. When a dropdown row's text is wider than the cell, the in-progress search needle inside the displayed value is replaced with an ellipsis so the surrounding context (record description, trailing key data) remains visible.

• Link-field dropdown now shows match count and search source. The dropdown popup carries a footer strip with two pieces of feedback: on the left, how many entries match against the total available (with a + when more results were truncated at the server fetch limit); on the right, a glyph indicating where the current results came from — the field's preloaded option list, the local search cache, or a live server lookup. Hovering either area shows a tooltip with details and the relevant keyboard shortcut. When the cache produces no matches, the footer replaces the counter with an explicit hint that a server search is one keystroke away.

• On-demand server search from the dropdown. Two new shortcuts let you reach past the local cache: Shift+Enter escalates a zero-match cache result to the server (so “no matches” doesn't quietly mean “not loaded yet”), and Ctrl+Enter discards the local cache entirely and forces a fresh server lookup (useful when you suspect the cache is stale). Both work from inside an inline cell editor without committing the cell.

• Right-click and middle-click on link-field dropdown rows. Right-clicking a row opens a context menu with Select, Cancel, Open in desktop, and Open in web client. The web-client entry follows the same per-class gating and prefer-web-client ordering already used by the main right-click menu, so administrators only need to configure web access in one place. Middle-clicking a row opens the inspector for that candidate directly without changing the linked record, and the dropdown stays open so you can keep browsing. Opening an inspector or switching to another application no longer dismisses the popup or loses the in-progress search — focus returning to the link field restores the popup at the same scroll position.

• Natural alphanumeric order in link-field dropdowns. When the filter doesn't narrow the cached options (empty search, or a search that matches every loaded row), entries are now sorted so that File 2 sorts before File 10 rather than between File 19 and File 20. Fields backed by a fixed option list retain their curated order.

• Optional relevance scoring for link-field search. A new Use relevance scoring for searching relations checkbox under Options → Misc ranks search candidates against a small set of signals — how recent the record's activity is, the position of certain option values (todo / SOS / history / document status), and whether the record is associated with the active user (responsible coworker, creator, or last editor). Off by default and stored per user. Intended for solutions where the obvious “best match” rarely sorts first alphabetically.

• Double-clicking a record to open it no longer causes a phantom click on the inspector that just appeared. With opens now fast enough that the inspector window can appear under the cursor before the user finishes releasing the second click, the trailing up of the double-click was leaking into the new window — focusing a field, dismissing a popover, or triggering whatever was under the cursor. The leak is now suppressed at the application level for all mouse buttons.

• Global search no longer fires against servers that do not support it. The global-search popup is now gated on whether the connected server actually has full-text search enabled, not just on whether the search-bar is visible.

• “Show super search on cards” setting controls inspector-card visibility again. A regression in earlier 11.3 builds caused the legacy super-search bar to appear on every inspector card regardless of the EnableInspectorFullTextSearch admin checkbox under Options → Misc. 11.2 behavior is restored: the bar shows on inspector cards only when the admin has opted in. The main window's bar follows the same capability check and stays hidden when neither the webclient integration nor server-side full-text search is available. Setting changes — the opt-in, the IntegrateWebClient / AllowWebClient toggles, or the server's full-text configuration — now reflect on every open inspector and the main window without needing to close and reopen anything. A new View → Bars → Search bar menu entry lets users hide the bar manually within the limits of the capability gate; the menu item is greyed out when no supplier is configured.

• Filters with OR-joined “not in” conditions now return the correct results. An internal simplification of expressions shaped like NOT IN A OR NOT IN B was producing the wrong set. The simplification is corrected.

• Inspector layout no longer flickers between expanded and collapsed states. A measurement bug in the auto-expand logic could leave the layout oscillating across re-layouts; the calculation is now font-relative and stable.

• Server-search shortcut no longer fires on link fields backed by a fixed option list. The local options list is complete by definition for these fields, so the Shift+Enter shortcut now no-ops instead of round-tripping to the server.

• TAB on “No matches” cancels instead of linking an invisible row. When typing in a relation field's search box filtered the popup to zero matches but the underlying preloaded options list was still populated, TAB would silently link the first entry in the unfiltered list — typically an option that didn't match what was typed and that the user never saw. TAB on a no-match popup now treats the action as Cancel (same as ESC and the right-click Cancel menu entry) and lets focus move normally to the next field.

• Prefix matches always rank above interior matches in relation search. When the typed text appears at the start of an entry, that entry now sorts above entries where the same text appears in the interior, regardless of the field's secondary sort (curated option-list order, logical alphanumeric, or cache relevance). Matches the user expectation that “what I typed should be at the top of the list”.

• Dropdown row text stays visible when the typed text matches the entire item. The long-row overflow elision (which substitutes the typed search text with an ellipsis to keep surrounding context visible) no longer collapses the row to a lone ellipsis when the entire item value equals the typed text. The row falls back to clipping at the cell boundary, so at least the prefix of the entry remains readable.

• Auto-generated relation-option queries skip virtual fields. When synthesising filter conditions from the source record's other relations, the path now skips link fields backed by SQL expressions (no physical column) and link fields without an in-table foreign key. Those edges can't participate in the synthesised join and previously caused either backend errors or filters that referenced the wrong column.

• Inactive-row filtering respects the “active” naming convention. Classes whose inactive-flag column is named active (value 0 = inactive) now compose the auto-include-inactive filter with the correct polarity. Classes using the legacy quit convention continue to work unchanged.

• Whitespace inside formula strings survives transpilation to VBScript. Lime-formula expressions containing significant whitespace inside string literals were being collapsed by the formula-to-VBScript lowering pass; the literals are now preserved verbatim across the lowering.

• Re-entrancy crash when multiple Lime links arrive close together. In CTI integrations that fire several limecrm:// links in rapid succession — call-center popup sequences in particular — a nested message pump inside one link's inspector creation could dispatch the next link's queued work, run it, and clear the shared queue out from under the outer iteration. The outer iterator then pointed into a cleared list, producing an access violation that the catch-all path reported as a generic “Internal error”. The receiver now snapshots its queue at entry, so nested re-entry can never invalidate the outer iteration; newly-arrived links stay on the queue and are drained on their own scheduled turn — nothing is lost.

• Nested field paths in filters no longer cache a partial resolution. Filter conditions that referenced a related-record field via a dotted path — for example, document.fileextension — could fail with a misleading “is not a valid value” error from the backend after the first attempt failed. The internal resolver was caching the partial path it had walked before the error, so subsequent calls saw the cached prefix as a successful resolution and produced a condition that referenced the wrong column. The resolver now caches only after the full path resolves cleanly; failures stay failures, and the next call retries against current schema state.

• Filters that combine multiple sub-filters no longer fail with “Condition field not found”. Some filter shapes that merge a sub-filter into the outer query — particularly those that reach across a relation into a nested expression — could fail at evaluation time because an internal join-scope path notation introduced during the merge was leaking into the field lookup. The lookup now strips that notation, so the previously-failing filters run as the user wrote them.

• Internal-error reports name the real source location. “An internal error has occurred in source file …” dialogs previously reported the same line (the line inside the catch wrapper) for every catch-all-handled exception, regardless of where the exception actually came from — useless for diagnosis. The message now identifies the actual call site of the catch that fired.

• Record.Value with mixed-case field names works again. A regression introduced earlier in 11.3 caused Record.Value(“xyz”) (or any other case variant of a field name) to silently return nothing once a different field had been looked up first in the same record. Mixed-case lookups now work consistently regardless of the order they're made in.

• Filter conditions for set-typed fields match whether the value is a single string or an array. Previously the same value resolved to different conditions depending on how it was passed: Field.Value = “Privat” bypassed the option-label lookup that Field.Value = Array(“Privat”) performed. Both forms now produce identical conditions.

• Column-filter checkmarks return for set-typed fields. Selecting a value in the column-filter dropdown for a set-typed field shows as checked again. The dropdown also de-duplicates correctly — re-clicking a checked entry removes the condition instead of adding a duplicate — and saved set-field column filters reload as proper conditions instead of being silently demoted to a text match.

• Decimal precision preserved when filtering. Filter values for decimal fields were silently truncated to integer before reaching the query, so fractional values were lost. Decimal values now flow through unchanged.

• Client no longer hangs when scripts modify records during AfterSave. A VBA AfterSave or BeforeSave handler writing back to the record while the underlying record load was still in flight could deadlock the client until it was force-killed. The locking and async-load sequence is now coordinated so the handler proceeds without blocking.

• Refresh in AfterSave handlers no longer deadlocks. A Controls.AfterSave handler calling Controls.Refresh() could deadlock against the still-running save flow. Refresh now defers correctly until the save flow exits and runs from there. A small set of related save-state tracking issues were addressed in the same pass.

• Ambiguous-record errors now identify the offending record. When the duplicate-row check trips during record load, the error now names the class and record ID so the entry can be located in the data. Definitive load errors that previously got absorbed into a generic warning are now surfaced to the user.

• Adding participants to a campaign no longer fails with “Item not found”. When the participant class had a virtual relationsingle field marked invisible-everywhere, the wizard's confirm step threw Item not found and aborted. The wizard now excludes virtual fields from its candidate field list — those fields were never assignable, so listing them was always a no-op (silent in 11.2, loud after 11.3's stricter field-list checks).

• Mass update auto-refreshes the table again. A regression vs 11.2 left the explorer showing pre-update values until manually refreshed. The Mass Update dialog now has a “Reload list after update” checkbox — default on, persisted per user — that requeries the explorer when the batch completes. The label notes that records can drop out of the visible list if their new value no longer matches the active filter.

• Inspector fields follow the splitter when scrolled to the bottom. Dragging the inspector's bottom splitter down with the field list scrolled to the end used to leave the fields anchored at their old screen positions with empty space below — the scrollbar range updated but the controls didn't move. They now reposition against the new scroll position so the bottom of the field list stays anchored to the bottom of the viewport.

• Stability when controls are detached. A null-pointer crash that could occur when fields fired before/after-command events on a control that had already been detached from its inspector is fixed. The previous symptom was a generic “internal error” log entry with no indication of cause.

• Avoided crashes from circular control configurations. When custom inspector layouts placed controls in mutually-referencing arrangements, an internal lookup could recurse and exhaust the stack. A guard now stops the recursion safely.

• Client-side incremental refresh reverted. The incremental refresh optimization shipped earlier in 11.3 produced unexpected load patterns on the backend in production. It is reverted in this build; refresh runs the full query as it did before the optimization. The optimization will be re-introduced once the backend impact is resolved.

• Trailing multi-line note fields fill the inspector again. Form definitions that used the legacy -1 “fill remaining” indicator on a multi-line field (such as Notes / Anteckningar in History inspectors) regressed to only two lines in earlier 11.3 builds. They now extend to the bottom of the inspector, matching 11.2 behavior. Short notes display without scrollbars; longer notes that need a scrollbar still get a usable three-line pane minimum even when the inspector is heavily compressed.

• AllowWebClient is opt-in again. A regression caused AllowWebClient to default to enabled in the Misc options page; opening Options → Misc and clicking OK silently persisted it as enabled. The dialog default now matches the runtime opt-in design. The setting also moved from per-Windows-user (HKCU) to database scope, so admins can manage it centrally and stale per-machine values from earlier 11.3 builds are ignored.

• Settings lookup tolerates a missing database. GetSettingInherited previously threw when Application.Database wasn't available - typical during early startup or after a disconnect - instead of falling through to the application-scope value. Missing database now means “no value at the database scope” and the lookup continues.

• Record.Copy / Record.CopyTo no longer moves or duplicates related records. Two distinct issues fixed: related entities reachable through any relation whose foreign-key column lives on the related side (virtual or not, single-link or multilink) were being reassigned to the new copy when saved, effectively moving them rather than copying — the source no longer loses its links. Separately, a 11.3-era deep clone branch for plain 1:1 link relations fabricated a phantom companion record on every copy; that branch is reverted to 11.2's no-op for the 1:1 case, so copying no longer creates extra related records the user didn't ask for.

• Inspector bottom splitter no longer locks in transient view states. Earlier 11.3 builds persisted the “fields collapsed” and “fields maximised” view states using sentinel values that were meant to be transient, so the inspector could come back stuck in one of those states on next launch. The persistence layer now ignores the legacy sentinels and emits values that round-trip cleanly.

• Lime CRM shows the correct icon in Windows Explorer. A new icon resource added during 11.3 shifted icon-resource ordering, causing Windows Explorer to pick a generic shape icon instead of the Lime logo for lime.exe. The original logo is restored. Users on machines where the icon cache had latched onto the wrong icon may need a one-time Explorer restart before the corrected icon shows up.

• Dim-fields-when-null is now opt-in. Following customer feedback, the visual dimming applied to fields whose value is NULL no longer defaults to enabled for new users. Users who actively enabled it keep their setting; users who never opened Options → Misc no longer see dimmed fields out of the box. The toggle is still available under Options → Misc.

• Update-field dialog targets the field you clicked, not a same-named system field. When a custom field's display name collided with a system field of the same label (e.g. a solution-defined Status alongside the built-in Status (System)), the dialog could open against the system field instead of the one the user picked from the column-header menu. The menu now passes the actual field reference instead of re-resolving by display name. The same fix applies to the explorer's view, filter, and Send-to menus, and to field pickers in the view-edit, filter-condition, and formula dialogs.

• Explorer no longer leaves empty placeholder columns for missing fields. When a view referenced fields that don't exist in the data (renamed, removed, or otherwise not loaded), the explorer kept empty column slots for them. Those slots desynced the column index used by sort and cell-rect math. Missing fields are now skipped completely so column-indexed operations stay aligned.

• Action Pane no longer blanks after hide-then-show. Hiding the inspector's Action Pane and showing it again could leave the pane visible but with the embedded browser blank. The host now explicitly re-attaches and repaints the browser when the pane becomes visible again.

• Inspector bottom splitter no longer restores at a sub-row sliver. Some startup paths (particularly the first card opened from a solution, where settings can race with VBA-driven record loading) could leave the splitter a few pixels from the top — the controls area effectively unusable. The splitter now applies its standard default position when no settings are available, and any persisted value smaller than a single layout row is treated as corrupt and falls back to the default.

• Maximize-tabs toolbar button reflects the effective state. The button's pressed-state and overlay glyph used to require the splitter to be exactly at its minimum position to register as “maximized”. Any controls area smaller than one layout row now counts — matching the user's perception that the controls area is gone.

• Maximize-tabs toolbar button restores to the default position. Clicking the button while tabs are maximized used to expand to the splitter's last drag-saved position, or to the midpoint if no drag had been saved. It now expands to the standard default position (tabs at the auto-expand limit from the bottom). The last drag position is still available via double-click on the splitter bar itself.

• Drag-drop save silently lost edits on the new record. Dropping a document onto an explorer item could create a record that the open path internally marked as a dummy placeholder. Subsequent saves silently no-op on dummy records — but BeforeSave and AfterSave events still fired, making the failure invisible. New records are no longer marked as dummies during the open path, attempted saves on a record that is a dummy now log as an error rather than silently no-opping, and any inspector that does end up bound to a dummy now opens read-only so accidental edits can't be lost.

• First / last / previous / next record navigation works again. The toolbar's record-navigation buttons could fail with Index out of range or a generic Invalid parameter error, with the inspector refusing to move to the requested record. The inspector's current-record position resolution is more robust against the case where the active item's stored position hasn't been renumbered against the explorer's sorted view, and a related validation step that was rejecting otherwise valid navigation targets has been corrected.

• Enter saves again when editing inline in the explorer. Earlier 11.3 builds inverted the Enter / Ctrl+Enter contract for inline cell editing — plain Enter moved to the next cell instead of saving the row, and customers reported it as the inline editor “essentially behaving like Tab”. 11.2 semantics are restored: Enter saves and closes, Ctrl+Enter moves to the next cell within the row, Ctrl+Shift+Enter moves backward. Multi-line text controls inside the cell continue to consume Enter and Shift+Enter for paragraph and soft-line breaks as before.

• Record shortcuts pasted into Outlook or webmail clients open again. Copying a record put two clipboard formats on the system clipboard, one of them an HTML payload containing a limecrm-shortcut link. The link inside that payload was URL-encoded a second time before being embedded, so a space arrived in the recipient's mail as %2520 instead of %20 and the link no longer resolved. The redundant second encoding is removed; the plain-text clipboard format was never affected.

• Scripts writing to mixed-case fields target the right field again. A regression caused by 11.3's case-insensitive Record.Value lookup let writes such as rec(“dealstatus”) = x fail with Status (System) is read-only on classes that had both a built-in status field and a user-defined dealstatus field — the lookup was crossing the two during cache resolution. The lookup now keys on each field's own local name, so writes go to the field the script names and read-only system fields no longer absorb the write.