Burada maxdop proqramının 24 avqust 2026 tarixindən bəri buraxılan 3 versiyası var; son versiya: 0.1.4. Hər versiya naşirin öz quraşdırma faylına aparır.
| Versiya | Tarix | Ölçü | Fayllar |
|---|---|---|---|
| 0.1.4Son versiya | 5 noyabr 2026 | — | 64 bit ZIPARM64 ZIP Versiya qeydləriNo formatting changes. A file formatted by 0.1.3 comes out of 0.1.4 byte for byte the same.
This release moves the parser forward and cleans up the toolchain behind it.
ScriptDom 180.140.0
maxdop parses with Microsoft's ScriptDom, and the pin moves from 180.102.0 to 180.140.0. The grammar
surface grew from 935 to 938 concrete node types, all three from one addition: the Fabric DW grammar's
scalar AI functions (AI_SUMMARIZE, AI_CLASSIFY, AI_TRANSLATE and the others) now take an
ON ERROR clause — ERROR ON ERROR, NULL ON ERROR or DEFAULT <value> ON ERROR.
Those functions exist only under the Fabric DW grammar; under every other grammar the same text is an
ordinary function call. maxdop has no layout rule for them and passes the whole expression through as
written, so the new clause needed no formatting work. What it got instead is a test: the clause
survives the round trip verbatim, inside a statement that still formats around it, and formatting the
result again is a fixed point.
The parser is the one thing in this release that can change what maxdop sees in a file, so it was
checked as such. The corpus sweep against 180.140.0 came back unchanged: nothing refused, nothing
crashed, nothing unstable.
Toolchain
- .NET SDK 10.0.403. The release binaries are Native AOT, so the .NET runtime is compiled into
them rather than installed beside them. The SDK pin moves from 10.0.302 to 10.0.403, which takes
the compiled-in runtime from 10.0.10 to 10.0.14 and with it the fix for CVE-2026-111497, an
out-of-bounds write in the runtime's Brotli decoder. maxdop never decodes Brotli, so it was not
reachable, but a binary that carries its own runtime should carry a patched one.
- VS Code extension. ovsx, the tool that publishes the extension to Open VSX, pulled in its own
older copy of @vscode/vsce beneath the one the extension already builds with, and with it a
dependency under a security advisory. An npm override now resolves ovsx to the top-level vsce,
which removes the duplicate tree and the advisory with it. This is build tooling: the extension's
own code is unchanged and it has no runtime npm dependencies. Separately, minimatch, which
vsce uses to apply .vscodeignore, is overridden to 11.0.1 or later for a denial-of-service
advisory (GHSA-g3hv-75wg-m249) that no vsce release has picked up yet.
- Test toolchain. Mutation testing runs on Stryker.NET 5.0, and the test projects move to
Microsoft.NET.Test.Sdk 18.11, coverlet 10.1 and xunit.runner.visualstudio 4.0.1. Pinned, as before,
so a mutation score or a coverage number stays comparable with last week's. |
| 0.1.2 | 31 avqust 2026 | — | 64 bit ZIPARM64 ZIP Versiya qeydləriCasing on Built-in Functions and Globals
Built-in function names and global variables now take keyword casing. Until now CAST, COALESCE, NULLIF, LEFT, RIGHT and IIF were recased while len and
row_number beside them were not, so a single expression could come out in two cases.
New Configuration
maxdop now recases from a list of built-in names, and that makes it the only casing decision in
the formatter not derived from the parse tree, which is controllable with a new configuration value:
{
...
"recaseBuiltInFunctions": true
}
It follows keywordCase, so "lower" brings GETDATE() and @@ROWCOUNT down rather than leaving
casing half-applied.
Setting it to false keeps the casing you wrote for everything the list covers — built-in function
names, global variables, the parser-matched table functions such as STRING_SPLIT, and the aggregate
a PIVOT names — so off means off rather than mostly off.
It does not reach casing that the parse tree proves. CAST, NVARCHAR, WITHIN GROUP and
DBCC CHECKDB … WITH NO_INFOMSGS stay cased under keywordCase with the switch off, because this
switch exists for the one decision made from a list instead of from the grammar.
Caveats:
- The call must have no qualifier. SQL Server requires at least a two-part name to invoke a
user-defined function, so an unqualified len(x) can only ever bind to the built-in. If you have a
function of your own called Len, you reach it as dbo.Len(x) — which has a call target and is
left exactly as you wrote it.
- The name must be undelimited. [len](x) is untouched. Brackets are how you say "this
identifier is spelled exactly like this".
- The list is consulted in function-name position only. A column named len, a table named
replace, an alias named getdate, a variable named @len. All plain names are untouched.
Global variables
IF @@ERROR <> 0 PRINT @@SERVERNAME; -- @@rowcount, @@trancount, @@fetch_status, …
DECLARE @@MyVar INT is legal T-SQL, and ScriptDom resolves a later reference by
spelling rather than by scope, so in an expression position a
local @@MyVar arrives as the very same GlobalVariableExpression that @@ROWCOUNT does. Recasing
on the prefix quietly renames that variable under a case-sensitive collation. So the
documented globals are listed by name, and a local variable spelled like a system one keeps every
character you wrote.
WITHIN GROUP is one clause
-- 0.1.1 -- 0.1.2
STRING_AGG(a, ',') within GROUP (ORDER BY b) STRING_AGG(a, ',') WITHIN GROUP (ORDER BY b)
GROUP is reserved and lexed as its own token; WITHIN is not reserved and lexed as an identifier.
The clause reached the output through a slice that recases neither names nor identifiers — because a
COLLATE Latin1_General_BIN can sit in that same region, and a collation name is a name. So half the
clause came up and half did not.
Fixed structurally rather than by spelling: WithinGroupClause is a node, so its head is a region the
grammar guarantees holds no name. The collation that shares the old region is not a node in the same
way, and keeps exactly the treatment it had. The clause also gained a layout, breaking at its own
parenthesis when the line is long, the same as OVER.
PIVOT aggregates
-- 0.1.1 -- 0.1.2
SELECT SUM(a) FROM t; SELECT SUM(a) FROM t;
… PIVOT (sum(amount) FOR m IN ([Jan])) p … PIVOT (SUM(amount) FOR m IN ([Jan])) p
The same split casing as WITHIN GROUP, in a different place. PIVOT names its aggregate through
AggregateFunctionIdentifier, which is a MultiPartIdentifier rather than a FunctionCall, so the
built-in list never saw it — a SUM in the select list came up while a sum two lines below did not.
DBCC
-- 0.1.1 -- 0.1.2
dbcc checkdb('MyDb') with no_infomsgs; DBCC CHECKDB('MyDb') WITH NO_INFOMSGS;
dbcc shrinkfile (MyFileName, 100); DBCC SHRINKFILE (MyFileName, 100);
Administrative statements reach the output through a generic fallback that normalises spacing and
descends into children but deliberately recases nothing, because it cannot tell a keyword from a name
in a construct nobody wrote a ha |
| 0.1.1 | 24 avqust 2026 | — | 64 bit ZIPARM64 ZIP |