"zero"/null of storable types, rework initializers #764
Conversation
| </xmp> | ||
| </div> | ||
|
|
||
| ## Storable types ## {#storable-types} |
There was a problem hiding this comment.
Moved this up above pointer types, since pointer types are not storable.
Fixes gpuweb#685 The generically zero-able types (or nullable types) correspond exactly with storable types, so I tied nullability to the storable type concept. Eliminate the concept of a "default" type. Instead define the null value of a storable type. Clarify that a variable declaration can have an initializer. Specify the initial value of a variable, and it's the null value for the store type if the declaration had no intializer. This still leaves up in the air exactly when a variable is "created" as we haven't specfiied an execution order or even the concept of execution of a statement. Initializers are only permitted for 3 storage classes Remove "Initializers" section later on, as its content is now mostly abosrbed into the "Variables" section. Hint at future reorganization, to insert a "Evaluation and Execution" section between "Types" and "Grammar" major sections.
|
This was a bit more sprawling a change than I had hoped. It touched variables and "default values" and add type rules. It feels like we're missing a big section in the middle to describe actual execution, something between Types and the Grammar. |
dneto0
added
the
wgsl resolved
| `mat3x4<f32>()` : mat3x4<f32><br> | ||
| `mat4x4<f32>()` : mat4x4<f32> | ||
| <td>Null value (OpConstantNull) | ||
| <tr> |
There was a problem hiding this comment.
Should we add struct and array in here as well for completeness?
There was a problem hiding this comment.
This table is only for matrix constructors.
There was a problem hiding this comment.
Sorry, I mean add a table for structs and arrays similar to the ones we have for other types add show the null value converts to OpConstantNull
|
|
I'd be ok with that, or with "nil". |
|
One thing to keep in mind is that, for shader languages that have C-like pointers, the all-zero-bits pointer into workgroup storage (local or shared) is often a valid address; the nullptr has non-zero bit pattern. But we're not there yet, and can leave that conceptual challenge to the future. |
|
Discussed at the 2020-05-19 meeting. Resolution was to use zero. |
Sometimes, the English seems to read better when writing "zero-valued" instead of just "zero".
|
I believe fall feedback is addressed. PTAL |
grunt-http-server hasn't been updated in 4 years and depends on a broken version of http-server that causes infinite redirects on Windows.
5 participants
Fixes #685
The generically zero-able types (or nullable types) correspond exactly with
storable types, so I tied nullability to the storable type concept.
Eliminate the concept of a "default" type. Instead define the null
value of a storable type.
Clarify that a variable declaration can have an initializer.
Specify the initial value of a variable, and it's the null value for
the store type if the declaration had no intializer.
This still leaves up in the air exactly when a variable is "created"
as we haven't specfiied an execution order or even the concept of
execution of a statement.
Initializers are only permitted for 3 storage classes
Remove "Initializers" section later on, as its content is now mostly
abosrbed into the "Variables" section.
Hint at future reorganization, to insert a "Evaluation and Execution"
section between "Types" and "Grammar" major sections.