Dawit Bekele

My assumption is that the dialect is unstated until proven otherwise, because SQL that runs on one server fails on another in ways that look like the query was wrong. The differences are small individually and endless in aggregate. Identifier quoting uses double quotes in the standard and backticks in MySQL, so a lowercase name quoted in one is an error or a case-mangled identifier in the other. Limiting rows is a limit clause in one, a top clause in another, and a fetch clause in a third.

String concatenation has an operator, a function and a concatenation operator depending on which one you are looking at. Date functions share names and nothing else. Formatting is more than cosmetics. A query formatted the way a tool prints it is often not the query a developer wrote, and the diffs produced by reformats bury the real change.

I format on a stable layout and treat a reformat-only commit as its own change. Validation is a separate step from formatting, and a formatter that only checks bracket balance will happily hand back a query with no FROM clause on a delete. I check structure and then explain what the statement does.

Minification is the reverse and rarely belongs in source. Squashing a query into one line makes it unreadable and saves nothing, since the query travels over a connection rather than a wire that charges by the byte. The performance section covers indexing and the join pattern worth avoiding: a query that runs once per returned row turns a page of results into hundreds of round trips. That one appears in more slow application code than anything else I see.