New naming convention proposal #149

Merged
Demeter Dzadik merged 12 commits from Mets/blender-studio-pipeline:mets-naming-conventions into main 2023-11-23 17:11:47 +01:00
Showing only changes of commit 7960580c16 - Show all commits

View File

@ -2,17 +2,18 @@
## Separators ## Separators
We only have 3 separators to work with, so we try keep their meaning consistent across conventions: We only have 3 separators to work with, so we try keep their meaning consistent across conventions:
`-` : Separates prefixes from each other and from the rest of the name. `-` : Separates prefixes (including the Asset Identifier) from each other and from the rest of the name.
`_` : Used instead of spacebar in composite words, eg. `eifel_tower_base`. `.` : Separates suffixes, currently only for symmetry sides, ie `.L`/`.R`.

I agree with this characterization of _ which to me stands in confilct of the proposal to use it for hierarchies/parts later in this document.

I agree with this characterization of `_` which to me stands in confilct of the proposal to use it for hierarchies/parts later in this document.

I think that later part of the document is over-explaining a bit, and makes things that I see as part of the base name (without any suffixes or prefixes) sound as if they were something technically significant. I'll try to clarify this better.

I think that later part of the document is over-explaining a bit, and makes things that I see as part of the base name (without any suffixes or prefixes) sound as if they were something technically significant. I'll try to clarify this better.
`.` : Suffixes, used only for symmetry sides, ie `.L`/`.R`, and for variations, `clock.normal`, `clock.broken`, `clock.destroyed` `_` : Used to separate parts of the base name, eg. `console_button_032`. The base name of an ID never has any technical significance.

To me these two categories are completely separate from each other. There could also be variations of symmetric objects.

I'd propose to use . generally for parts/hierarchies like we have in the past. .L/R is compatible with that definition imo.
Variations are for me something orthogonal to that. Could we introduce :? Not sure why we are limited to those 3, as compatibility shouldn't be an issue.

To me these two categories are completely separate from each other. There could also be variations of symmetric objects. I'd propose to use `.` generally for parts/hierarchies like we have in the past. `.L/R` is compatible with that definition imo. Variations are for me something orthogonal to that. Could we introduce `:`? Not sure why we are limited to those 3, as compatibility shouldn't be an issue.

I agree with introducing : as a separator for variants, I only didn't do this because I wasn't sure if you guys are open to it.

Only note is that I would still always put .L/.R at the very end, even if there is a variation, otherwise blender's built-in name flipping function wouldn't work. (Which to be fair isn't used for any built-in operators in Object mode, but for add-ons it's still handy.)

I agree with introducing `:` as a separator for variants, I only didn't do this because I wasn't sure if you guys are open to it. Only note is that I would still always put `.L/.R` at the very end, even if there is a variation, otherwise blender's built-in name flipping function wouldn't work. (Which to be fair isn't used for any built-in operators in Object mode, but for add-ons it's still handy.)
`:` : Separator for variations of an object, eg. `clock:normal`, `clock:broken`, `clock:destroyed`.

I had to google what mnemonic means. Maybe we can move to something more intuitive like identifier?

I had to google what mnemonic means. Maybe we can move to something more intuitive like identifier?

Agreed!

Agreed!
## Asset Identifier ## Asset Identifier
All local datablocks across all assets of a production must have a unique name. To faciliate this, each asset is assigned an identifier that must remain unique within a given production. For example, a character named "Elder Sprite" might get the identifier "ES". The max length of this identifier is not limited, but the shorter the better. If a minor prop in the production is called "Electric Switch", it can not have the identifier "ES", because it's already taken. So, "ELSW" would be fine. All local datablocks across all assets of a production must have a unique name. To faciliate this, each asset's name, or a shortened version of it, should be present in all ID names of that asset. For example, the character "Elder Sprite" might have a full-length unique identifier of `elder_sprite`, along with an optional shortened (but still unique) version, eg. `esprite`.

The mnemonics should have more 4 letters (more than the 2 and 3) to distinguish them from asset prefixes. otherwise it would be easy to confuse the two. maybe making them lower case would help too.

The mnemonics should have more 4 letters (more than the 2 and 3) to distinguish them from asset prefixes. otherwise it would be easy to confuse the two. maybe making them lower case would help too.

If we don't mind longer ID names, I'm happy to use just a slightly shortened version of names, eg. "elder" for Elder Sprite or "rocket_int" for Rocket Interior.

If we don't mind longer ID names, I'm happy to use just a slightly shortened version of names, eg. "elder" for Elder Sprite or "rocket_int" for Rocket Interior.
This identifier will be used in the names of datablocks, to make sure all names are unique, not just in a single file, but the whole production. This is useful/necessary because of the way Blender's Override system works, which relies on making local copies of linked object hierarchies, meaning all assets need to be able to exist in the same namespace without collisions. This identifier will be present among the prefixes in the names of datablocks, to make sure all names are unique, not just in a single file, but across the whole production. This is useful/necessary because of the way Blender's Override system works, which relies on making local copies of linked object hierarchies, meaning all assets need to be able to exist in the same namespace without name collisions.
## Asset Collections ## Asset Collection Hierarchy
We use prefixes for the root collections of assets to help distinguish different types of assets in the Outliner of a complex shot file. The name of the asset itself should be lowercase. We use prefixes for the root collections of assets to help distinguish different types of assets in the Outliner of a complex shot file. The name of the asset itself should be lowercase.
@ -27,59 +28,55 @@ Example: `CH-elder_sprite`
Note that there's no technical distinction between different types of assets. This is purely for organizational purposes and comfort. Note that there's no technical distinction between different types of assets. This is purely for organizational purposes and comfort.

I don't remember this being a necessity in the current design of the asset pipeline. Maybe I'm misremembering but I don't generally think that binding the tasklayer association with the name is a good idea, since that will make changing this without breakage impossible.

I don't remember this being a necessity in the current design of the asset pipeline. Maybe I'm misremembering but I don't generally think that binding the tasklayer association with the name is a good idea, since that will make changing this without breakage impossible.

It is, it's necessary for each task layer to be assigned its own piece of the hierarchy, otherwise the pipeline doesn't know which hierarchy to keep; The one being pulled in, or the one already in the current file. So when pulling into Rigging currently, it will preserve the current hierarchy for the Rigging collection, but take the hierarchy of the other collections from the publish.

It is, it's necessary for each task layer to be assigned its own piece of the hierarchy, otherwise the pipeline doesn't know which hierarchy to keep; The one being pulled in, or the one already in the current file. So when pulling into Rigging currently, it will preserve the current hierarchy for the Rigging collection, but take the hierarchy of the other collections from the publish.

But I thought we agreed on storing this data as custom properties and keeping the hierarchies out of it, no?
Because this isn't just about the exceptions, right? Every collection would have that. Having the tasklayer name in every single collection of an asset seems quite off to me, that's not something I remember considering. Maybe that was a misunderstanding at the time.

But I thought we agreed on storing this data as custom properties and keeping the hierarchies out of it, no? Because this isn't just about the exceptions, right? Every collection would have that. Having the tasklayer name in every single collection of an asset seems quite off to me, that's not something I remember considering. Maybe that was a misunderstanding at the time.

So as far as the Asset Pipeline is concerned, we did agree on storing ownership information inside property groups on the individual objects, and that is how ownership of objects is handled by the Asset Pipeline.

But for consistency in the organization of collections each task layer has a collection, the user can assign objects to their task layer collection. These top-level collections are named after the task layers and are purely organizational and don't influence who owns what object, but they are necessary to avoid conflicts in collection assignments during Push/Pull. Personally I prefer the collection set-up like this, but I'm open to discussing alternatives.

It is possible to replace this feature with some kind of ownership system and UI similar to objects for collections, but that not within the scope of things we are currently focused on. We are more focused on finalizing the transfer logic and getting 'ownership' working on non-object IDs like GeoNode groups.

So as far as the Asset Pipeline is concerned, we did agree on storing ownership information inside property groups on the individual objects, and that is how ownership of objects is handled by the Asset Pipeline. But for consistency in the organization of collections each task layer has a collection, the user can assign objects to their task layer collection. These top-level collections are named after the task layers and are purely organizational and don't influence who owns what object, but they are necessary to avoid conflicts in collection assignments during Push/Pull. Personally I prefer the collection set-up like this, but I'm open to discussing alternatives. It is possible to replace this feature with some kind of ownership system and UI similar to objects for collections, but that not within the scope of things we are currently focused on. We are more focused on finalizing the transfer logic and getting 'ownership' working on non-object IDs like GeoNode groups.

It's actually not necessary for the sub-collections' names to include the task layer name, that's true. That was more of a stylistic choice on my end. So you're right, this is in fact legal:

CH-elder_sprite
    esprite-model
        esprite-head
        esprite-eyes
            esprite-eye.L
            esprite-eye.R
    esprite-rig
    esprite-shading

I think it might be a bit nicer with the task layer names, when seeing the collections as a flat list, like when appending or using the Blend File view of the Outliner. Also because multiple task layers might have their own collection relating to a given part of a character. But I don't mind too much if this isn't strictly enforced.

It's actually not necessary for the sub-collections' names to include the task layer name, that's true. That was more of a stylistic choice on my end. So you're right, this is in fact legal: ``` CH-elder_sprite esprite-model esprite-head esprite-eyes esprite-eye.L esprite-eye.R esprite-rig esprite-shading ``` I think it might be a bit nicer with the task layer names, when seeing the collections as a flat list, like when appending or using the Blend File view of the Outliner. Also because multiple task layers might have their own collection relating to a given part of a character. But I don't mind too much if this isn't strictly enforced.
The immediate sub-collections of the root collection are strictly defined by our Asset Pipeline add-on, as being `{identifier}-{task_layer}`. Task Layers are the different data layers that make up an asset, such as Modeling, Rigging, and Shading. Inside these Task Layer Collections, each artist responsible for their own Task Layer may create sub-collections freely, as long as each sub-collection still starts with `{identifier}-{task_layer}-`. The immediate sub-collections of the root collection are strictly defined by our Asset Pipeline add-on, as being `{identifier}-{task layer}`. Task Layers are the different data layers that make up an asset, such as Modeling, Rigging, and Shading. Inside these Task Layer Collections, each artist responsible for their own Task Layer may create sub-collections freely, as long as each sub-collection still starts with `{identifier}-{task layer}-`.
So, here's what an asset's collection hierarchy might look like: So, here's what an asset's collection hierarchy might look like:
```txt ```txt
CH-elder_sprite CH-elder_sprite
Review

I find it quite confusing if the main asset collection prefix and any other collection prefix share the exact same formatting. I'd propose to keep those clearly different to avoid confusion and improve readability.

So either making the identifier not capitalized or using a different separator or both.

I find it quite confusing if the main asset collection prefix and any other collection prefix share the exact same formatting. I'd propose to keep those clearly different to avoid confusion and improve readability. So either making the identifier not capitalized or using a different separator or both.
Review

This seems to align with Andy's suggestion of keeping these identifiers longer. In that case, I agree with lower-case, so instead of "ES" or "elder_sprite", we could just have "esprite" or "elder" (up to whoever is first creating the asset). For most assets, especially characters, our names tend to be quite short, so additional shortening isn't always needed. Only downside is that maybe sometimes we'll end up with some pretty long names, but it's not the end of the world.

This seems to align with Andy's suggestion of keeping these identifiers longer. In that case, I agree with lower-case, so instead of "ES" or "elder_sprite", we could just have "esprite" or "elder" (up to whoever is first creating the asset). For most assets, especially characters, our names tend to be quite short, so additional shortening isn't always needed. Only downside is that maybe sometimes we'll end up with some pretty long names, but it's not the end of the world.
ES-model esprite-model
ES-model-head esprite-model-head
ES-model-teeth_and_tongue esprite-model-teeth_and_tongue
ES-rig esprite-rig
ES-rig-widgets esprite-rig-widgets
ES-rig-helpers esprite-rig-helpers
ES-rig-eye.L esprite-rig-eye.L
ES-rig-eye.R esprite-rig-eye.R
ES-shading esprite-shading
ES-shading-outlines esprite-shading-outlines
``` ```
## Asset Datablocks ## Name Prefixes
- All local datablocks of an asset (Object, Mesh, Material, Action, etc.) must include either the asset's identifier or full name (both is not necessary). Eg., `RIG-elder_sprite` or `GEO-ES-eye.L`. This is automated by the Asset Pipeline add-on. All Object names (not just in Assets) must start with a prefix describing the object's purpose:
- `LGT` : Light objects and mesh-lights, also shadow casters
- `ENV` : Matte paintings, sky-domes, fog volume objects
- `GEO` : Geometry, meshes, curves that contribute to the rendered appearance of the asset
- `GPL` : Grease pencil stroke objects (need to differentiate from GEO because can not be rendered on the farm)
- `RIG` : Rig and rig specific objects that do not appear in rendering
- `WGT` : Bone shapes
- `HLP` : Empties and other helper objects that are not rendered

This needs stronger wording than "should be advisable", every object datablock should have that prefix to make it clear when linking/appending what that object is and to visually distinguish the objects in the outliner and python query.

This needs stronger wording than "should be advisable", every object datablock should have that prefix to make it clear when linking/appending what that object is *and* to visually distinguish the objects in the outliner and python query.

Gotcha, will replace "should" with "must".

Gotcha, will replace "should" with "must".
- `TMP` : Any object used in pre-viz that should be replaced with final assets over the course of the production.
All local datablocks of an asset (Object, Mesh, Material, Action, etc.) must also include either the asset's shortened or full name. Eg., `RIG-elder_sprite` or `GEO-elder-eye.L`. This is enforced by the Asset Pipeline add-on.
## Base Names
- Object Data and Shape Keys should be named the same as the containing Object. This is automated by the Asset Pipeline add-on. - Object Data and Shape Keys should be named the same as the containing Object. This is automated by the Asset Pipeline add-on.
- Datablock names must not end with a `.00x` suffix. This is enforced by the Asset Pipeline add-on. - Words in all base names should be lower-case and separated by `_`.
- When there's too many objects to manually name, like when building a house out of a hundred wooden plank objects, the **Batch Rename Datablocks** built-in add-on should be used to give groups of objects the same name. Then replace the `.` in the `.00x` suffixes with an `_` instead, so `GEO-HS-wooden_plank.023` becomes `GEO-HS-wooden_plank_023`. - All datablock names must not end with a `.00x` suffix. This is enforced by the Asset Pipeline add-on.
- When there's too many objects to manually name, like when building a house out of a hundred wooden plank objects, the **Batch Rename Datablocks** built-in add-on should be used to give groups of objects the same name. Then replace the `.` in the `.00x` suffixes with an `_` instead, so `GEO-HS-wooden_plank.023` becomes `GEO-HS-wooden_plank_023`, making the number a part of the base name.
- The purpose of this is that when there are >1 copies of this asset overridden (and therefore local) in a file, every object in the same copy of the asset will have the same number suffix. The first copy will have no suffix. The second copy of the asset will have object names like `GEO-HS-wooden_plank_023.001`, which is fine. If we didn't do this step, this object would instead end up getting named `GEO-HS-wooden_plank.024`, which is not fine because now the suffix number has no correlation with which overridden copy of the asset this object belongs to. - The purpose of this is that when there are >1 copies of this asset overridden (and therefore local) in a file, every object in the same copy of the asset will have the same number suffix. The first copy will have no suffix. The second copy of the asset will have object names like `GEO-HS-wooden_plank_023.001`, which is fine. If we didn't do this step, this object would instead end up getting named `GEO-HS-wooden_plank.024`, which is not fine because now the suffix number has no correlation with which overridden copy of the asset this object belongs to.
- Without this step, debugging shot files with multiple copies of an asset becomes a nightmare, and it could even confuse the Library Override system and break shots. - Without this step, debugging shot files with multiple copies of an asset becomes a nightmare, and it could even confuse the Library Override system and break shots.
It is advisable to give Objects an additional prefix to provide information about the object's purpose, especially when it is NOT part of an asset. This is purely for organizational and comfort purposes. Here are some recommendations: ## Name Suffixes
Use `.L`/`.R` suffixes for objects that belong to one side and are symmetrical.

according to the first section, _ is used only as spacebar replacement, it's not a separator. so an of relationship could be - instead

according to the first section, `_` is used only as spacebar replacement, it's not a separator. so an _of_ relationship could be `-` instead

Thinking about it now, I actually don't think an of relationship needs to be signified by the ID name at all, since it has no technical significance in the pipeline. To me, it's just the same as any other word separation. Collections also help with this.

Thinking about it now, I actually don't think an *of* relationship needs to be signified by the ID name at all, since it has no technical significance in the pipeline. To me, it's just the same as any other word separation. Collections also help with this.
- `WGT` : Bone shapes
- `LGT` : Light objects and mesh-lights, also shadow casters
- `HLP` : Empties and other helper objects that are not rendered
- `GEO` : Geometry, meshes, curves that contribute to the rendered appearance of the asset
- `RIG` : Rig and rig specific objects that do not appear in rendering
- `ENV` : Matte paintings, sky-domes, fog volume objects
- `GPL` : Grease pencil stroke objects (need to differentiate from GEO because can not be rendered on the farm)
- `TMP` : Any object used in pre-viz that should be replaced with final assets over the course of the production.
Use `.L`/`.R` suffices for objects that belong to one side and are symmetrical.
On the other hand, avoid using `.L`/`.R` when similar objects exist on each side of an asset but aren't meant to be symmetrical. On the other hand, avoid using `.L`/`.R` when similar objects exist on each side of an asset but aren't meant to be symmetrical.
Example: `GEO-WRDB-drawer_knob.L` for the left-side drawer knob of a wardrobe prop. "Of" relationships should be part of the base name, and not a suffix. Example: `GEO-wardrobe-drawer_knob.L` for the left-side drawer knob of the Wardrobe prop.
If a name contains an "*of*" relationship - in the above example the `knob` *of* the `drawer` - these should NOT be separated by `.`, but rather with `_`. If the watch had a variant of type `clean` and `dirty`, these would be using a `:` to express the nature of the variant:
Another example:
``` ```
Good: `GEO-ES-watch_screw` `GEO-watch.clean`

dot . should become a dash -
everything else looks good and can be merged as far as I am concerned :)

dot `.` should become a dash `-` everything else looks good and can be merged as far as I am concerned :)

Thanks, fixed!

Thanks, fixed!
Bad: `GEO-ES-watch.screw` `GEO-watch.dirty`
```
If the watch had a variant of type `clean` and `dirty`, these would be using a `.` to express the nature of the variant:
```
`GEO-ES-watch.clean`
`GEO-ES-watch.dirty`
``` ```
## Actions ## Actions
@ -90,15 +87,15 @@ Pose Library actions are in their own files, and marked as assets, so they show
Examples: Examples:
``` ```
Hand_Fist hand_fist
Hand_Splayed hand_splayed
Face_Happy face_happy
Face_Surprised face_surprised

we established the "no caps, no gaps" rule at some point, should be used here as well. so no uppercase letters except for prefix and suffixes.

we established the "no caps, no gaps" rule at some point, should be used here as well. so no uppercase letters except for prefix and suffixes.
Eyes_Blink eyes_blink
Eyes_Surprised eyes_surprised
``` ```
### Rig Constraints ### Rig Actions
Rigs may use Actions for Action Constraint based control set-ups. Since these are asset datablocks, they must start with the asset identifier. Rigs may use Actions for Action Constraint based control set-ups. Since these are asset datablocks, they must start with the asset identifier.
Corrective Actions are ones which are meant to activate when two other actions activate. These should use the name of the two trigger actions, separated by a "+". Corrective Actions are ones which are meant to activate when two other actions activate. These should use the name of the two trigger actions, separated by a "+".
@ -118,3 +115,25 @@ ANI-ES-060_scratch_layout.v001
ANI-ES-110_0100_A.v001 ANI-ES-110_0100_A.v001
ANI-REX-140_0020_A.v001 ANI-REX-140_0020_A.v001
``` ```
### Bone Names
Making rigs as easy to learn and work with as possible is the most important thing when naming bones.
Every bone name should be named properly (never `Bone.023`), avoid ugly number suffixes, and avoid unnecessary padding in numbers.
Controls that are exposed to animators should ideally have max one prefix, with an intuitive acronym.
Examples:
```
Bad: STR-TIP-Finger_04_003.L
Good: STR-Finger_Pinky_4.L
```
Numbering should start from 1, but if a bone gets inserted at the start of the chain, it's better to just number it 0 rather than change all the pre-existing bones' names, to not break any existing animations.
Symmetry sides should be indicated by `.L`/`.R`.
Other prefixes and their meaning:
- `IK` : Control body parts in Inverse Kinematics mode
- `FK` : Control body parts in Forward Kinematics mode
- `STR` : Squash and stretch body parts
- `ROOT` : Control a cohesive area of the rig (often detachable)
- `TGT` : Control the target of a Look-At rig set-up
- `P` : Parent of another control