Quote predicate names as identifiers where fragment quoting renders broken SQL

[simon-mundy]

Problem

A predicate's left-hand name is rendered through AbstractPlatform::quoteIdentifierInFragment(). It splits the name on every character outside the platform's word class and quotes each piece, except spaces, dots, * and AS. A name that holds any other delimiter therefore renders as a different, invalid identifier:

Call	Today (SQL-92)	After
where->equalTo('x-y', 1)	"x""-""y" = '1'	"x-y" = '1'
where(['t-1.x-y' => 1])	"t""-""1"."x""-""y" = '1'	"t-1"."x-y" = '1'
where->like('a"b', …)	"a""'""b" LIKE …	"a'b" LIKE …
where->isNull('nâme')	"n""\xC3""\xA2""me" IS NULL (split bytes)	"nâme" IS NULL
having->greaterThan('my col', 1)	"my" "col" > '1'	"my col" > '1'

This affects equalTo, notEqualTo, lessThan(OrEqualTo), greaterThan(OrEqualTo), like/notLike, in/notIn, between/notBetween, isNull/isNotNull and where-array keys, in WHERE and HAVING on Select, and in WHERE on Update and Delete. It has the same root cause as #238 (Update SET columns) and #241 (Select, join and GROUP BY columns).

Cause

AbstractSql::processExpression() renders every Argument\Identifier through quoteIdentifierInFragment(), and so does processIdentifiersArgument() for Argument\Identifiers. Fragment quoting is meant for loose SQL, not for a single name.

The non-breaking rule

On 0.6.x only output that is broken today may change. A new private AbstractSql::quoteIdentifierArgument() decides per name:

• Split the name on space, . and *. Treat ., * and AS as operators.
• Fragment quoting is valid when every remaining name part comes out of the platform's quoteIdentifierInFragment() as one piece (it equals quoteIdentifier() of that part), and no two name parts are separated by spaces alone.
• If fragment quoting is valid, the output is today's, byte for byte.
• Otherwise the name is quoted whole as an identifier chain: split only on ., with each part through quoteIdentifier() and a * part left bare. This matches the helpers in Quote Update SET columns as identifiers and bind them under valid placeholder names #238 and Quote Select, join and GROUP BY column names as identifiers #241.

The check asks the platform itself, so a platform whose fragment pattern treats - as a word character, as phpdb-mysql's does, keeps `a-b` * `c` unchanged.

What stays the same

• Plain names, dotted names, t.*, price*qty, a * b, t.a * t.b, x as y / x AS y, a,b and the empty string render exactly as before.
• Expression, Literal and string predicates are untouched, and so are values and string join ON clauses (they call quoteIdentifierInFragment() directly with extra safe words).
• No public or protected API changes.

The same fix also reaches the other Identifier arguments rendered through processExpression(): identifier-typed Expression parameters and DDL names (column, index and constraint names). For those too, only names that render broken today change.

Context

The maintainer's long-term direction (D18) is that identifier slots take identifiers only, with SQL passed as an Expression. That change breaks forms like price*qty, so it belongs in a later major version. This PR fixes only what is broken today.

Tests

test/unit/Sql/PredicateIdentifierNameTest.php (#[Group('unit')]) uses data providers to cover:

• Broken forms now quoted as one identifier: hyphen, space, quote character, non-ASCII, tab, dotted hyphen, dotted space and hyphenated qualified star. Each runs across equalTo, notEqualTo, lessThan, greaterThan, like, in, between and isNull, where-array keys with a value and with null, HAVING, Update WHERE and Delete WHERE.
• Valid forms proven byte-identical across the same paths: plain, dotted, t.*, price*qty, a * b, t.a * t.b, x as y, x AS y, a,b and empty.
• An Argument::identifiers() list, with each name quoted by the same rule.
• A platform with a MySQL-style fragment pattern (new TrustingHyphenWordPlatform test asset), which keeps hyphenated names and arithmetic as before.
• Expression, Literal and string predicates left unchanged.

The changed lines have 100% line coverage. The unit suite passes, and mago format --check, mago lint and mago analyze are clean.

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.

📢 Thoughts on this report? Let us know!