Don't parse all attributes as Meta

[NyxCode]

Currently, sqlx parses every field attribute as Meta.
This breaks for every struct which has non-meta attributes with custom syntax. In ormx, for example, I have attributes which may have a path as value, for example: #[ormx(table = crate::models::User)].

When parsing attributes, I suggest we first check if the first token is sqlx, then continue parsing as Meta.

[dragonnn]

diff --git a/sqlx-macros/src/derives/attributes.rs b/sqlx-macros/src/derives/attributes.rs
index 3feea50b..4d2dadbb 100644
--- a/sqlx-macros/src/derives/attributes.rs
+++ b/sqlx-macros/src/derives/attributes.rs
@@ -123,13 +123,13 @@ pub fn parse_child_attributes(input: &[Attribute]) -> syn::Result<SqlxChildAttri
     let mut rename = None;
     let mut default = false;
 
-    for attr in input {
+    for attr in input.iter().filter(|a| a.path.is_ident("sqlx")) {
         let meta = attr
             .parse_meta()
             .map_err(|e| syn::Error::new_spanned(attr, e))?;
 
         match meta {
-            Meta::List(list) if list.path.is_ident("sqlx") => {
+            Meta::List(list) => {
                 for value in list.nested.iter() {
                     match value {
                         NestedMeta::Meta(meta) => match meta {

I think something like that should fix that problem, right? I can not test at the moment since I have problem build my project with a local version of sqlx. Complains about that no runtime is enabled (but it is).

[NyxCode]

Yeah, that should do the trick!

[jplatte]

I wonder what the map_err is there for. Either way, the second check you added it redundant but this should fix the issue 👍

[dragonnn]

I removed the second check when you look at the patch. Ok, I will create a PR in a minute :)

[jplatte]

Oh, right! 😅

To make it easier to read diffs in GitHub comments you can start the codeblock with ```diff 😉