High compile time of crates using sqlparser(codegen phase) - any way to reduce generated code?​​

[echou]

In my project, some crates have very long compilation times (e.g., 40-50 seconds, with the codegen phase accounting for 70%-90% of that time). I suspect these crates are slow to compile due to their dependency on sqlparser.

I analyzed the compilation of one such crate using cargo llvm-lines. The top functions by lines are listed below:

Lines                  Copies               Function name
  -----                  ------               -------------
  2180708                31532                (TOTAL)
   220326 (10.1%, 10.1%)    21 (0.1%,  0.1%)  <sqlparser::ast::Statement as sqlparser::ast::visitor::VisitMut>::visit
   163970 (7.5%, 17.6%)   1995 (6.3%,  6.4%)  <alloc::vec::Vec<T> as sqlparser::ast::visitor::VisitMut>::visit
   117960 (5.4%, 23.0%)   3654 (11.6%, 18.0%) <core::option::Option<T> as sqlparser::ast::visitor::VisitMut>::visit
    76507 (3.5%, 26.5%)     21 (0.1%, 18.0%)  <sqlparser::ast::Expr as sqlparser::ast::visitor::VisitMut>::visit

As shown, the visitor trait implementations for the AST types generate a significant amount of code. Is there any way to reduce the code generation footprint of sqlparser?

[LucaCappelletti94]

Hi, which compilation flags are you using? I assume you are using the visitor-generated code, otherwise a trivial solution is not to enable that feature. How much does the compilation time change with and without the visitor feature enabled?

In my own projects, where admittedly I have no use of the visitor feature, the time requirement of compiling sqlparser is rather negligible.

[echou]

Yes, my project uses visitor feature heavily. The crates not using sqlparser are quite fast to compile.

[LucaCappelletti94]

There was recently an update in the sqlparser derive crate - have you already tried whether the new version has improved / worsened these time requirements?