Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Replace NoiseModelFactor1 in with NoiseModelFactorN in pre-made factors #1344

Merged
merged 4 commits into from
Dec 23, 2022

Conversation

gchenfc
Copy link
Member

@gchenfc gchenfc commented Dec 23, 2022

This is a follow-up to #947 which applies this "new style" NoiseModelFactorN to existing factors in GTSAM (e.g. PriorFactor, BetweenFactor, ...) which, after #947 is merged, they will be using deprecated versions of NoiseModelFactor123456.

I decided to separate this into a separate PR to keep file changes cleaner and because I think it might warrant more discussion (*sigh) and I don't want to drag out the other one if it doesn't have to.

Hiccup

In the process, I realized that this will introduce some backward compatibility issues. Specifically:

  • For new usages / new factors, everything is fine.
  • But for existing factors, once we change them from inheriting from NoiseModelFactor2 (deprecated) to NoiseModelFactorN, users will lose access to the key2() functions and the ::X2 type aliases (inherited from NoiseModelFactor2) since the new usage is now key<2>() and ::ValueType<2>.

So, for example (I will use this example for the rest of this description, but all pre-made factors in GTSAM have similar issues):

template <typename T>
class BetweenFactor : public NoiseModelFactor2<T, T> { ... }; // <-- change to NoiseModelFactorN<T, T>

BetweenFactor<Pose3> factor(X(1), X(2), Pose3(), noise);
EXPECT( X(1), factor.key1() ); // <-- key1() needs to change to key<1>()
EXPECT( X(2), factor.key2() ); // <-- key2() needs to change to key<2>()
using T1 = typename BetweenFactor<Pose3>::X1; // <-- `::X1` needs to change to `::ValueType<1>`
static_assert(std::is_same<T1, Pose3>::value, "X1 type incorrect");

Proposed Solutions

Allowing backward compatible access to key2() is not too ugly, so I just moved those into NoiseModelFactorN and deprecated them (fully backward compatible, but deprecated).

But allowing backward compatible access to ::X2 is quite ugly (see here and below), so I just left these out (implemented in NoiseModelFactor2, but not accessible from BetweenFactor anymore). I see 3 most reasonable options:

  1. Break backward compatibility and, if anyone uses BetweenFactor::X2, they should change to BetweenFactor::ValueType<2>. Use of ::X2 seems pretty uncommon and didn't appear ever inside gtsam codebase including examples & tests (all unit tests pass).
  2. Implement deprecated versions of X1 and X2 inside BetweenFactor (and for every pre-made factor in GTSAM). This is fine, but will just take a while and might add a lot of clutter.
  3. Implement the ugly version in NoiseModelFactorN (see below).

template <class... VALUES>
class NoiseModelFactorN
: public NoiseModelFactor,
public detail::AliasX<VALUES...>, // using X = VALUE1
public detail::AliasX1<VALUES...>, // using X1 = VALUE1
public detail::AliasX2<VALUES...>, // using X2 = VALUE2
public detail::AliasX3<VALUES...>, // using X3 = VALUE3
public detail::AliasX4<VALUES...>, // using X4 = VALUE4
public detail::AliasX5<VALUES...>, // using X5 = VALUE5
public detail::AliasX6<VALUES...> // using X6 = VALUE6
{

Changes / notes:

  • NoiseModelFactor123456 -> NoiseModelFactorN
  • BACKWARD COMPATIBILTY BREAKAGE: (see above) previously X2 is now ValueType<2>. e.g. any instances of BetweenFactor<Pose3>::X2 need to be changed to BetweenFactor::ValueType<2>. This applies to all pre-made factors in GTSAM.
  • move the deprecated key1() functions from NoiseModelFactor1 to NoiseModelFactorN. Although we had previously discussed that it would be cleaner to be in NoiseModelFactor1, this will avoid breaking backward compatibility when changing pre-made factors to inherit from NoiseModelFactorN.
  • Thing I recently noticed: when we have a templated factor class that inherits from NoiseModelFactorN, some calls (inside the class only) get ugly due to dependent types. e.g. template keyword is required here:
    << keyFormatter(this->template key<1>()) << ","
    << keyFormatter(this->template key<2>()) << ")\n";

    This only affects calls inside the class and is similar to why we need to sometimes use typename. More details here. I know this is standard template syntax, but I do remember a couple years ago the first time I saw this syntax reading code I was really confused because it looks so weird, so I would like to minimize its use if possible to avoid confusing people.

@gchenfc gchenfc requested a review from dellaert December 23, 2022 06:28
@dellaert
Copy link
Member

@gchenfc , I like solution 2 the best: just add using X1 = in the factors within GTSAM. Also, for some reason, this PR did not change its base to develop when I merged #947 , so you’ll have to change it manually…

@varunagrawal varunagrawal changed the base branch from feature/NoiseModelFactorN to develop December 23, 2022 16:03
@varunagrawal
Copy link
Collaborator

I was wondering for a second why am I suddenly getting so many deprecation warnings. 😅

@varunagrawal
Copy link
Collaborator

Can we land this quickly please? Debugging is hard with the crazy amount of deprecation messages...

@dellaert
Copy link
Member

Can we land this quickly please? Debugging is hard with the crazy amount of deprecation messages...

Yeah, I screwed up, did not see the other PR was targeting #947

Copy link
Member

@dellaert dellaert left a comment

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I will approve so we can land to suppress the warnings, but I did not see
“solution 2” implemented anywhere. Is it no longer needed?

@dellaert
Copy link
Member

I will let @gchenfc merge - maybe solution 2 is still in the works, although I would also support doing this in a different PR.

@gchenfc
Copy link
Member Author

gchenfc commented Dec 23, 2022

Sounds good, yes solution 2 can be a separate PR. Might not be able to get to it until tomorrow night, but this PR can stand on its own before solution 2.

Merging now.

@gchenfc gchenfc merged commit df5b894 into develop Dec 23, 2022
@gchenfc gchenfc deleted the feature/NoiseModelFactorN_replaceDeprecated branch January 5, 2023 03:57
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
None yet
Projects
None yet
Development

Successfully merging this pull request may close these issues.

3 participants