[WIP] Mandatory degree #2097
[WIP] Mandatory degree #2097
Conversation
georgwiese
force-pushed
the
mandatory-degree
branch
from
b7f9d7d
to
00a7bd5
Compare
| @@ -22,16 +22,16 @@ mod utils { | |||
|
|
|||
| machine Main with degree: 8 { | |||
| // use a machine from another module by relative path | |||
| my_module::Other a; | |||
| my_module::Other a(8, 8); | |||
There was a problem hiding this comment.
This started as a one-day-before-the-realease-hack and now it's mandatory?
There was a problem hiding this comment.
In my understanding, the hacky part is the syntax around how degrees are passed to sub-machines. I don't think that we want to pass degree ranges is controversial?
There was a problem hiding this comment.
For me it is, shouldn't it somehow at least have a default? What do I care about the degree of a sub-machine, unless I already know that I will call it every single row like the memory machine?
There was a problem hiding this comment.
I think it's hard to pick a good default heuristically. It depends on the block size and how often we expect to call the machine, and both the user would know better than us.
There was a problem hiding this comment.
Basically,
- The status quo (pre Remove env var MAX_DEGREE_LOG #2092) is that the default is
min_degree: 2**5, max_degree: 2**MAX_DEGREE_LOG, whereMAX_DEGREE_LOGis an environment variable. But using env variables is weird. - With Remove env var MAX_DEGREE_LOG #2092, we fail if the degree range is unknown after the linker. This is already kind of making it mandatory to set the range, except that we have a strange rule in the linker: If the degree of the main machine is static, use that degree for all sub-machines (even if it's already been set differently). That rule is also weird and should be removed IMO. It also doesn't have any effect if the main machine is dynamically sized, which should be the case in most real-world applications.
There was a problem hiding this comment.
The problem I see with having default ones is that only very large defaults are safe enough, but with that you make the life of any user that doesn't need it worse because fixed columns gen and setup time will be massive
There was a problem hiding this comment.
Why not generate the fixed columns only once they are needed? Because of the verification key?
There was a problem hiding this comment.
Yes, exactly, we only know what's needed at runtime, but at that time, we already need to have committed to the fixed columns in all possible sizes.
Idea: After machine instantiation the type of the degree changes from
MachineDegree(optional min / max) toNamspaceDegree(mandatory min / max).Requires more changes in the test: in many tests, we rely on a to-be-removed linker feature (if the main degree is static, it is used for all sub-machines). Instead, we should always set the degree range at instantiation time.