[Form lib] Correctly add field to form on component mount #75796

sebelga · 2020-08-24T17:03:51Z

This PR fixes #74677. A regression was introduced from the work of memorizing the field and form objects and handlers in (#71237).

Security solution forms

Fixing the form lib to correctly add and remove fields to the form hook has revealed some bugs in the security solution plugin. Mainly in the <StepAboutRuleComponent />.

I discovered that there were nested form fields which are not allowed in the form lib. There was a riskScore field and a riskScore.value field. In the form library, fields correspond to the leaves of an object.

// There are 3 fields for the `riskScore` object. But the object **is not** a form field
const formData = {
  riskScore: {
    value: 'hello', // field = 'riskScore.value'
    isMappingChecked: true, // field = 'riskScore.isMappingChecked'
    mapping: [] // field = 'riskScore.mapping' (This should probably be a <UseArray /> field
  }
}

I fixed the issue by removing the riskScore.value field.
I found the same issue with the security level and also fixed it.

Lesson learned

The lesson learned from the regression introduced in the lib is the importance of the order of the useEffect on the file.

// This was how it was with the bug

// [1]
const onValueChange = useCallback(async () => {
  ...
  await form.validateFields([path]); // Validate the current field (passing its "path")
}, [value, <other-deps>]);

// [2]
const field = useMemo(() => {
  return {
    value,
    ...
  }
}, [value, <other-deps>]);

useEffect(() => {
  if (!isMounted.current){
    return;
  }
  onValueChange(); // [4]
}, [onValueChange])

useEffect() => {
  form.addField(field); // [5]
  
  return () => {
    form.removeField(path); // [3]
  }
}, [field, path]);

In the above snippets, we can see that when the field value changes it

Creates a new callback onValueChange
Creates a new field object
Remove the field from the form (!)
Call onValueChange() which in turn asks to validate the field
Add the field back to the form

So on step 4, it asks to validate the field but the field was removed on step 3! So nothing to validate === bug.

The solution is simple, but it took me a while to figure it out! 😢

// This was how it was before

// [1]
const onValueChange = useCallback(async () => {
  ...
  await form.validateFields([path]); // the current field that has changed
}, [value, <other-deps>])

// [2]
const field = useMemo(() => {
  return {
    value,
    ...
  }
}, [value, <other-deps>]


useEffect() => {
  form.addField(field); // [4]
  
  return () => {
    form.removeField(path); // [3]
  }
}, [field, path]);

useEffect(() => {
  if (!isMounted.current){
    return;
  }
  onValueChange(); // [5] // The field has already been added so no more problem!
}, [onValueChange])

…orm object from field I removed the form object form the field object to avoid unnecessary effect triggers. The form object should be accessed by context with the "useFormContext()" hook.

…ounted

…ield-on-component-mount

sebelga · 2020-08-25T14:57:29Z

.../plugins/security_solution/public/detections/components/rules/step_about_rule/index.test.tsx

-        ],
-      };
-      expect(stepDataMock.mock.calls[1][1]).toEqual(expected);
+    await act(async () => {


Note: by wrapping the button click event inside the await act, we don't need the waitFor call below. This allows us to get a proper error from the test instead of a jest timeout.

That should be the only use of waitFor in this file.

Yes it seems so but I didn't touch the other tests that also wrap the expect inside a waitFor. This test was the one failing and the timeout error did not let me see why it was failing.

Oh, absolutely! I just meant that the import is now unused.
Edit: And I apologize, because I am wrong! 🤦

elasticmachine · 2020-08-25T15:24:38Z

Pinging @elastic/es-ui (Team:Elasticsearch UI)

sebelga · 2020-08-25T17:06:46Z

@elasticmachine merge upstream

…:sebelga/kibana into bug/form-lib-addfield-on-component-mount

…ield-on-component-mount

sebelga · 2020-08-26T09:43:01Z

@elasticmachine merge upstream

…:sebelga/kibana into bug/form-lib-addfield-on-component-mount

The way the information about a rule is shown depends on keys defined on the form schema. This is a potential future bug as there is only 1 field declared in the JSX (e.g. "riskScore") so what is inside the object is expected to be FieldConfig properties.

jloleysens

Great work @sebelga ! I tested locally with the my Grok form processor component that uses UseArray and the reordering works as expected with these changes (unfortunately don't have a PR up for this, yet).

Did not test mappings or security form changes.

jloleysens · 2020-08-27T07:34:45Z

src/plugins/es_ui_shared/static/forms/hook_form_lib/hooks/use_form.ts

+
+      const areAllFieldsValidated = formFieldsValidity.every(({ 1: isValidated }) => isValidated);
+
+      // If *not* all the fiels have been validated, the validity of the form is unknown, thus still "undefined"


Suggested change

// If *not* all the fiels have been validated, the validity of the form is unknown, thus still "undefined"

// If *not* all the fields have been validated, the validity of the form is unknown, thus still "undefined"

jloleysens · 2020-08-27T07:38:21Z

src/plugins/es_ui_shared/static/forms/hook_form_lib/hooks/use_form.ts

+
+      // At this stage we have an updated field validation state inside the "validationResultByPath" object.
+      // The fields we have in our "fieldsRefs.current" have not been updated yet with the new validation state
+      // (isValid, isValidated...) as this will happen _after_, when the "useEffect" triggers and calls "addField()".


After reading this I had a look at the type annotations for the FormHook interface. Do you think you could add some docs to each of those fields to explain intended use? I see __addField is called inside of useField hook and I am assuming that is the call site you are referring to here?

Yes I could do that. Indeed I was referring to __addField. Although all those handlers starting with a double underscore are not meant to be public and thus I wouldn't want to have a doc appear on top of their suggestion in the IDE.

sebelga · 2020-08-27T08:35:25Z

Thanks for the review @jloleysens ! I keep a note about adding some doc on top of the interface properties. I will do that along with some other improvement of TS on the lib that I have in mind.

sebelga · 2020-08-27T08:35:35Z

@elasticmachine merge upstream

patrykkopycinski

Thank you @sebelga 💪

kibanamachine · 2020-08-27T10:28:36Z

💚 Build Succeeded

continuous-integration/kibana-ci/pull-request
Commit: b0c271f

Build metrics

async chunks size

id	value	diff	baseline
indexManagement	1.6MB	+439.0B	1.6MB
ingestPipelines	705.5KB	+22.0B	705.5KB
securitySolution	9.5MB	+467.0B	9.5MB
total		+928.0B

page load bundle size

id	value	diff	baseline
esUiShared	990.6KB	+1.4KB	989.2KB

History

💔 Build #70513 failed 44b5572
💔 Build #70498 failed c529e80
💔 Build #70479 failed 20e723d
💔 Build #70458 failed ef8aa16
💔 Build #70447 failed d04098b

To update your PR or re-run it, just comment with:
@elasticmachine merge upstream

sebelga · 2020-08-27T12:02:44Z

Thanks for the review @patrykkopycinski!

…) (#76077) Co-authored-by: Elastic Machine <elasticmachine@users.noreply.github.com> Co-authored-by: Elastic Machine <elasticmachine@users.noreply.github.com>

* Removes redundant uses of field's defaultValue * Most are redundant with the form's defaultValue; added calls to field.setValue in cases where the default is generated internally * Removes calls to reset() after submitting * These were needed due to a bug in the form lib; once elastic#75796 is merged these will no longer be necessary.

* Remove unused isNew field Digging through the git history, it looks like this was replaced with the isUpdateView prop at some point. There's a small chance that we're indirectly leveraging the effect of this value being changed, but I think we're safe. * WIP: Making rule form type safe We have lots of anys and unknowns in here, this is my attempt to fix that. I started by defining better types for the state/refs in our parent component; everything else mostly flowed out of that: * Step components now type their form hook for their step's data * Removes lots of unneeded `as` casts * Renames uses of `accordionId` with `step` and `activeStep`, since they are also values of our `RuleStep` enum * Step components now export their default values * The data flow is simpler when the parent passes these values in, rather than trying to merge props with some internal defaults. * The internal defaulting is still there, but I think it'll be unnecessary once I audit the `edit` forms. I've only done this work for the Define step for now, the rest are next. * Make defaultValues a required prop of the define step Now that the create step is passing in the default values, we no longer need to merge with internal state. The one exception is the default indexes; since we need that data for our "reset to default indexes" behavior, we'll keep that functionality within our DefineStep component. * Refactor rule creation forms to not require default values We don't gain much by forcing the parent to pass in default values. The slightly cleaner types are not worth the burden to the parent; instead, we add a type guard to be used in our parent to ensure our values are present before working with them. Previously, we were circumventing this logic with an `as` cast. * Remove unnecessary "deep" comparison These are arrays of strings, so a shallow comparison should suffice here. Also reorders conditions to short-circuit on simple booleans first. * Make StepRuleDescription generic on its schema * Fixes bug introduced by form lib updates There's currently a bug on master where returning to a previous form step does not populate its previous values. After some investigation it appears that this is due to form values being reset on submission (form.reset()). Previously, we kept a separate copy of data in each step's state, and had a useEffect that would repopulate the form's values if they ever became out of sync. Once that was removed I believe this bug was introduced. For now the fix is effectively reimplementing the above behavior, albeit a little more elegantly with `reset()`. If we can restructure the form logic to only require the form data at the end (big if), then we can remove the need to "repopulate" these values to the form. For now, this does remove the local copy of data in the step component as I believe that is no longer needed. Data is stored in the parent, copied/modified to the form, and pushed back up when one clicks on to the next step. * Rename typed hook to obviate eslint exception The linter was complaining because it didn't think that `typedUseKibana` was a hook. But it is, and we should name it as such. * WIP: Fixing type errors in the other form steps Things still aren't quite working, state gets lost when moving through steps but I believe this is addressed in an outstanding PR so I'm not sweating it right now. * Removes as much state in Step components as possible * We shouldn't need this as the form holds all the state as well. If we need to "watch" for a change, we can subscribe to the form's observable to replace FormDataProvider and local state (TODO) * Removes setting of default values in form components * I believe that this is redundant with defaultValues provided to useForm, but I need to verify. * More form cleanup * Removes redundant uses of field's defaultValue * Most are redundant with the form's defaultValue; added calls to field.setValue in cases where the default is generated internally * Removes calls to reset() after submitting * These were needed due to a bug in the form lib; once #75796 is merged these will no longer be necessary. * Fix some leftover type errors * Remove duplicated useEffect hook This exists identically earlier in the component; I'm guessing it was the result of a bad merge conflict. * Fix Rule edit form * Makes data structures more similar to rule creation form * Adds type guards for asserting which step is "active" * Simplifies logic around the active tabe/step/form * Fixes About Step jest tests * Removes use of wait() in favor of act() * Fixes mock call assertion now that we're no longer setting our data to null (which was a now-unnecessary form lib workaround). * Fix bug with going to a previous step after editing actions We never send our actions data back down to the actions form, so it was lost if you went to a previous step. Since the actions UI still had any connectors you created, you merely had to reselect the throttle and the connector, but this prevents you from having to do that. * Add assertions to our rule creation test Asserts that our rule form repopulates with the provided values when going back to a previous step. This is to cover a regression that was not caught by CI (but which has now been fixed). * Simplify Rule Creation logic * Validation and data collection are performed in the parent, not the step component * Step component provides a form ref and notifies the parent when it's being submitted; the rest is the parent's responsibility * Renames some internal helper functions to be more declarative: submitStep, editStep, etc. * Don't persist empty form data when leaving a form step If the active step form is invalid we will receive no data, so we must not persist that lest the form blows up on absent values when we later navigate back. * Skip About Step tests for now These exercise functionality that was moved into the parent, so they need a new home. * Remove unnecessary calls to setValue * Instead of setting our kibana url after the form is created, we add it to the form's default state * We do not need to set the throttle field value, the field component already does this * Style: logic cleanup * Prevent users from navigating away from an invalid step on rule edit We can go against the form lib conventions and persist this invalid data ourselves on transition, but for now this brings the create/edit forms into alignment so that adding the aforementioned behavior should be nearly identical on both. * Display callout if attempting to navigate away from an invalid tab We already do this if you try to _submit_ the form, but not when you click on another tab. * Persist our form submit() rather than the entire form Making the entire stateful form a useEffect dependency was likely causing unnecessary render cycles. This may also have been part of why both the hooks and the data are refs instead of normal state; to prevent these rerenders. * Replace FormDataProvider with useFormData hook We have to do a type cast here because the hook's data is not typed, but other than that this is a solid win and cleans things up immensely; the side effects that result from these values changing are much more apparent (IMO). * Move fetch of fields data _after_ form initialization This ensures that our first fetch of fields will use the index patterns on the form, not necessarily the default ones. * Replace FormDataProvider on About step * This fixes a bug where changing the default severity no longer updated the default risk score. It looks like this was broken when the severity/riskScore overrides were added, and the values of these fields changed from primitives to objects. * Replace local state with useFormData By watching the value directly from the form we no longer have a need for local state, as we were just using it to determine whether the throttle had changed from the default. * Types cleanup Remove some unneeded casts, add some needed ones. * Rewrite About Step tests Rather than asserting that the form is invalid through the UI, we instead retrieve data out of the form hook and assert on that instead. * Add memoization back to StepRuleDescription I'm not sure that it's necessary, but best to leave it until we have time to audit/remove multiple of these. * Do not fetch ML Jobs if StepRuleDescription is not rendering ML Data We were incorrectly invoking the useSecurityJobs hook any time the StepRuleDescription component was rendered, regardless of whether we actually needed that data. This moves the useSecurityJobs hook into the component that uses it, MlJobDescription. If we end up having multiple of these on the page we can evaluate caching/sharing this data somehow, but for now this prevents: * 3x3=9 unnecessary ML calls on the Rule Create page. * 1x3=3 unnecessary ML calls on Rule Details * 1x3=3 unnecessary ML Calls on the Rule Edit page. * Fix bug where revisiting the About step could modify the user's Risk Score With the severity/risk score link back in place, there was a bug where a user who had previously set a manual risk score would have it rewritten on edit (or edit during rule creation). This was due to a poorly-written useEffect that basically said "if there is a severity, set a risk score." This has now been amended to say "if the user changes the severity, set a risk score." * Clean up About Step tests * We don't need act(), it's not doing anything. * We don't need to click Continue since we're just talking to the form hook * Fix local form data when form isn't mounted If the form isn't on the page (e.g. if we're read-only), then useFormData will return no values. In these cases, we can simply fall back to the initialState values, as they'll either be: the default values on a new form, or: the current values on an active create/edit form. Updates the manual type of useFormData to reflect this "maybe" fact. * Allow user to navigate between invalid tabs on Edit Rule * Form hooks now _always_ return the form's data, regardless of validity * Edit Rule now: * persists invalid data * submits both the active form and the destination form on navigation. This is necessary to refresh validations on the destination form, since the form lib assumes a newly-mounted form is valid * simplifies "invalid tab" logic to be derived from our persisted data * Fix logical error If the rule is immutable, they can only edit actions. * Remove unneeded eslint exception Fixed by upstream #76471 * Make 21 the default risk score for a new rule Since the default severity is 'low,' these two defaults now coincide. * Remove duplicated type in favor of common one

* Remove unused isNew field Digging through the git history, it looks like this was replaced with the isUpdateView prop at some point. There's a small chance that we're indirectly leveraging the effect of this value being changed, but I think we're safe. * WIP: Making rule form type safe We have lots of anys and unknowns in here, this is my attempt to fix that. I started by defining better types for the state/refs in our parent component; everything else mostly flowed out of that: * Step components now type their form hook for their step's data * Removes lots of unneeded `as` casts * Renames uses of `accordionId` with `step` and `activeStep`, since they are also values of our `RuleStep` enum * Step components now export their default values * The data flow is simpler when the parent passes these values in, rather than trying to merge props with some internal defaults. * The internal defaulting is still there, but I think it'll be unnecessary once I audit the `edit` forms. I've only done this work for the Define step for now, the rest are next. * Make defaultValues a required prop of the define step Now that the create step is passing in the default values, we no longer need to merge with internal state. The one exception is the default indexes; since we need that data for our "reset to default indexes" behavior, we'll keep that functionality within our DefineStep component. * Refactor rule creation forms to not require default values We don't gain much by forcing the parent to pass in default values. The slightly cleaner types are not worth the burden to the parent; instead, we add a type guard to be used in our parent to ensure our values are present before working with them. Previously, we were circumventing this logic with an `as` cast. * Remove unnecessary "deep" comparison These are arrays of strings, so a shallow comparison should suffice here. Also reorders conditions to short-circuit on simple booleans first. * Make StepRuleDescription generic on its schema * Fixes bug introduced by form lib updates There's currently a bug on master where returning to a previous form step does not populate its previous values. After some investigation it appears that this is due to form values being reset on submission (form.reset()). Previously, we kept a separate copy of data in each step's state, and had a useEffect that would repopulate the form's values if they ever became out of sync. Once that was removed I believe this bug was introduced. For now the fix is effectively reimplementing the above behavior, albeit a little more elegantly with `reset()`. If we can restructure the form logic to only require the form data at the end (big if), then we can remove the need to "repopulate" these values to the form. For now, this does remove the local copy of data in the step component as I believe that is no longer needed. Data is stored in the parent, copied/modified to the form, and pushed back up when one clicks on to the next step. * Rename typed hook to obviate eslint exception The linter was complaining because it didn't think that `typedUseKibana` was a hook. But it is, and we should name it as such. * WIP: Fixing type errors in the other form steps Things still aren't quite working, state gets lost when moving through steps but I believe this is addressed in an outstanding PR so I'm not sweating it right now. * Removes as much state in Step components as possible * We shouldn't need this as the form holds all the state as well. If we need to "watch" for a change, we can subscribe to the form's observable to replace FormDataProvider and local state (TODO) * Removes setting of default values in form components * I believe that this is redundant with defaultValues provided to useForm, but I need to verify. * More form cleanup * Removes redundant uses of field's defaultValue * Most are redundant with the form's defaultValue; added calls to field.setValue in cases where the default is generated internally * Removes calls to reset() after submitting * These were needed due to a bug in the form lib; once elastic#75796 is merged these will no longer be necessary. * Fix some leftover type errors * Remove duplicated useEffect hook This exists identically earlier in the component; I'm guessing it was the result of a bad merge conflict. * Fix Rule edit form * Makes data structures more similar to rule creation form * Adds type guards for asserting which step is "active" * Simplifies logic around the active tabe/step/form * Fixes About Step jest tests * Removes use of wait() in favor of act() * Fixes mock call assertion now that we're no longer setting our data to null (which was a now-unnecessary form lib workaround). * Fix bug with going to a previous step after editing actions We never send our actions data back down to the actions form, so it was lost if you went to a previous step. Since the actions UI still had any connectors you created, you merely had to reselect the throttle and the connector, but this prevents you from having to do that. * Add assertions to our rule creation test Asserts that our rule form repopulates with the provided values when going back to a previous step. This is to cover a regression that was not caught by CI (but which has now been fixed). * Simplify Rule Creation logic * Validation and data collection are performed in the parent, not the step component * Step component provides a form ref and notifies the parent when it's being submitted; the rest is the parent's responsibility * Renames some internal helper functions to be more declarative: submitStep, editStep, etc. * Don't persist empty form data when leaving a form step If the active step form is invalid we will receive no data, so we must not persist that lest the form blows up on absent values when we later navigate back. * Skip About Step tests for now These exercise functionality that was moved into the parent, so they need a new home. * Remove unnecessary calls to setValue * Instead of setting our kibana url after the form is created, we add it to the form's default state * We do not need to set the throttle field value, the field component already does this * Style: logic cleanup * Prevent users from navigating away from an invalid step on rule edit We can go against the form lib conventions and persist this invalid data ourselves on transition, but for now this brings the create/edit forms into alignment so that adding the aforementioned behavior should be nearly identical on both. * Display callout if attempting to navigate away from an invalid tab We already do this if you try to _submit_ the form, but not when you click on another tab. * Persist our form submit() rather than the entire form Making the entire stateful form a useEffect dependency was likely causing unnecessary render cycles. This may also have been part of why both the hooks and the data are refs instead of normal state; to prevent these rerenders. * Replace FormDataProvider with useFormData hook We have to do a type cast here because the hook's data is not typed, but other than that this is a solid win and cleans things up immensely; the side effects that result from these values changing are much more apparent (IMO). * Move fetch of fields data _after_ form initialization This ensures that our first fetch of fields will use the index patterns on the form, not necessarily the default ones. * Replace FormDataProvider on About step * This fixes a bug where changing the default severity no longer updated the default risk score. It looks like this was broken when the severity/riskScore overrides were added, and the values of these fields changed from primitives to objects. * Replace local state with useFormData By watching the value directly from the form we no longer have a need for local state, as we were just using it to determine whether the throttle had changed from the default. * Types cleanup Remove some unneeded casts, add some needed ones. * Rewrite About Step tests Rather than asserting that the form is invalid through the UI, we instead retrieve data out of the form hook and assert on that instead. * Add memoization back to StepRuleDescription I'm not sure that it's necessary, but best to leave it until we have time to audit/remove multiple of these. * Do not fetch ML Jobs if StepRuleDescription is not rendering ML Data We were incorrectly invoking the useSecurityJobs hook any time the StepRuleDescription component was rendered, regardless of whether we actually needed that data. This moves the useSecurityJobs hook into the component that uses it, MlJobDescription. If we end up having multiple of these on the page we can evaluate caching/sharing this data somehow, but for now this prevents: * 3x3=9 unnecessary ML calls on the Rule Create page. * 1x3=3 unnecessary ML calls on Rule Details * 1x3=3 unnecessary ML Calls on the Rule Edit page. * Fix bug where revisiting the About step could modify the user's Risk Score With the severity/risk score link back in place, there was a bug where a user who had previously set a manual risk score would have it rewritten on edit (or edit during rule creation). This was due to a poorly-written useEffect that basically said "if there is a severity, set a risk score." This has now been amended to say "if the user changes the severity, set a risk score." * Clean up About Step tests * We don't need act(), it's not doing anything. * We don't need to click Continue since we're just talking to the form hook * Fix local form data when form isn't mounted If the form isn't on the page (e.g. if we're read-only), then useFormData will return no values. In these cases, we can simply fall back to the initialState values, as they'll either be: the default values on a new form, or: the current values on an active create/edit form. Updates the manual type of useFormData to reflect this "maybe" fact. * Allow user to navigate between invalid tabs on Edit Rule * Form hooks now _always_ return the form's data, regardless of validity * Edit Rule now: * persists invalid data * submits both the active form and the destination form on navigation. This is necessary to refresh validations on the destination form, since the form lib assumes a newly-mounted form is valid * simplifies "invalid tab" logic to be derived from our persisted data * Fix logical error If the rule is immutable, they can only edit actions. * Remove unneeded eslint exception Fixed by upstream elastic#76471 * Make 21 the default risk score for a new rule Since the default severity is 'low,' these two defaults now coincide. * Remove duplicated type in favor of common one

* Remove unused isNew field Digging through the git history, it looks like this was replaced with the isUpdateView prop at some point. There's a small chance that we're indirectly leveraging the effect of this value being changed, but I think we're safe. * WIP: Making rule form type safe We have lots of anys and unknowns in here, this is my attempt to fix that. I started by defining better types for the state/refs in our parent component; everything else mostly flowed out of that: * Step components now type their form hook for their step's data * Removes lots of unneeded `as` casts * Renames uses of `accordionId` with `step` and `activeStep`, since they are also values of our `RuleStep` enum * Step components now export their default values * The data flow is simpler when the parent passes these values in, rather than trying to merge props with some internal defaults. * The internal defaulting is still there, but I think it'll be unnecessary once I audit the `edit` forms. I've only done this work for the Define step for now, the rest are next. * Make defaultValues a required prop of the define step Now that the create step is passing in the default values, we no longer need to merge with internal state. The one exception is the default indexes; since we need that data for our "reset to default indexes" behavior, we'll keep that functionality within our DefineStep component. * Refactor rule creation forms to not require default values We don't gain much by forcing the parent to pass in default values. The slightly cleaner types are not worth the burden to the parent; instead, we add a type guard to be used in our parent to ensure our values are present before working with them. Previously, we were circumventing this logic with an `as` cast. * Remove unnecessary "deep" comparison These are arrays of strings, so a shallow comparison should suffice here. Also reorders conditions to short-circuit on simple booleans first. * Make StepRuleDescription generic on its schema * Fixes bug introduced by form lib updates There's currently a bug on master where returning to a previous form step does not populate its previous values. After some investigation it appears that this is due to form values being reset on submission (form.reset()). Previously, we kept a separate copy of data in each step's state, and had a useEffect that would repopulate the form's values if they ever became out of sync. Once that was removed I believe this bug was introduced. For now the fix is effectively reimplementing the above behavior, albeit a little more elegantly with `reset()`. If we can restructure the form logic to only require the form data at the end (big if), then we can remove the need to "repopulate" these values to the form. For now, this does remove the local copy of data in the step component as I believe that is no longer needed. Data is stored in the parent, copied/modified to the form, and pushed back up when one clicks on to the next step. * Rename typed hook to obviate eslint exception The linter was complaining because it didn't think that `typedUseKibana` was a hook. But it is, and we should name it as such. * WIP: Fixing type errors in the other form steps Things still aren't quite working, state gets lost when moving through steps but I believe this is addressed in an outstanding PR so I'm not sweating it right now. * Removes as much state in Step components as possible * We shouldn't need this as the form holds all the state as well. If we need to "watch" for a change, we can subscribe to the form's observable to replace FormDataProvider and local state (TODO) * Removes setting of default values in form components * I believe that this is redundant with defaultValues provided to useForm, but I need to verify. * More form cleanup * Removes redundant uses of field's defaultValue * Most are redundant with the form's defaultValue; added calls to field.setValue in cases where the default is generated internally * Removes calls to reset() after submitting * These were needed due to a bug in the form lib; once #75796 is merged these will no longer be necessary. * Fix some leftover type errors * Remove duplicated useEffect hook This exists identically earlier in the component; I'm guessing it was the result of a bad merge conflict. * Fix Rule edit form * Makes data structures more similar to rule creation form * Adds type guards for asserting which step is "active" * Simplifies logic around the active tabe/step/form * Fixes About Step jest tests * Removes use of wait() in favor of act() * Fixes mock call assertion now that we're no longer setting our data to null (which was a now-unnecessary form lib workaround). * Fix bug with going to a previous step after editing actions We never send our actions data back down to the actions form, so it was lost if you went to a previous step. Since the actions UI still had any connectors you created, you merely had to reselect the throttle and the connector, but this prevents you from having to do that. * Add assertions to our rule creation test Asserts that our rule form repopulates with the provided values when going back to a previous step. This is to cover a regression that was not caught by CI (but which has now been fixed). * Simplify Rule Creation logic * Validation and data collection are performed in the parent, not the step component * Step component provides a form ref and notifies the parent when it's being submitted; the rest is the parent's responsibility * Renames some internal helper functions to be more declarative: submitStep, editStep, etc. * Don't persist empty form data when leaving a form step If the active step form is invalid we will receive no data, so we must not persist that lest the form blows up on absent values when we later navigate back. * Skip About Step tests for now These exercise functionality that was moved into the parent, so they need a new home. * Remove unnecessary calls to setValue * Instead of setting our kibana url after the form is created, we add it to the form's default state * We do not need to set the throttle field value, the field component already does this * Style: logic cleanup * Prevent users from navigating away from an invalid step on rule edit We can go against the form lib conventions and persist this invalid data ourselves on transition, but for now this brings the create/edit forms into alignment so that adding the aforementioned behavior should be nearly identical on both. * Display callout if attempting to navigate away from an invalid tab We already do this if you try to _submit_ the form, but not when you click on another tab. * Persist our form submit() rather than the entire form Making the entire stateful form a useEffect dependency was likely causing unnecessary render cycles. This may also have been part of why both the hooks and the data are refs instead of normal state; to prevent these rerenders. * Replace FormDataProvider with useFormData hook We have to do a type cast here because the hook's data is not typed, but other than that this is a solid win and cleans things up immensely; the side effects that result from these values changing are much more apparent (IMO). * Move fetch of fields data _after_ form initialization This ensures that our first fetch of fields will use the index patterns on the form, not necessarily the default ones. * Replace FormDataProvider on About step * This fixes a bug where changing the default severity no longer updated the default risk score. It looks like this was broken when the severity/riskScore overrides were added, and the values of these fields changed from primitives to objects. * Replace local state with useFormData By watching the value directly from the form we no longer have a need for local state, as we were just using it to determine whether the throttle had changed from the default. * Types cleanup Remove some unneeded casts, add some needed ones. * Rewrite About Step tests Rather than asserting that the form is invalid through the UI, we instead retrieve data out of the form hook and assert on that instead. * Add memoization back to StepRuleDescription I'm not sure that it's necessary, but best to leave it until we have time to audit/remove multiple of these. * Do not fetch ML Jobs if StepRuleDescription is not rendering ML Data We were incorrectly invoking the useSecurityJobs hook any time the StepRuleDescription component was rendered, regardless of whether we actually needed that data. This moves the useSecurityJobs hook into the component that uses it, MlJobDescription. If we end up having multiple of these on the page we can evaluate caching/sharing this data somehow, but for now this prevents: * 3x3=9 unnecessary ML calls on the Rule Create page. * 1x3=3 unnecessary ML calls on Rule Details * 1x3=3 unnecessary ML Calls on the Rule Edit page. * Fix bug where revisiting the About step could modify the user's Risk Score With the severity/risk score link back in place, there was a bug where a user who had previously set a manual risk score would have it rewritten on edit (or edit during rule creation). This was due to a poorly-written useEffect that basically said "if there is a severity, set a risk score." This has now been amended to say "if the user changes the severity, set a risk score." * Clean up About Step tests * We don't need act(), it's not doing anything. * We don't need to click Continue since we're just talking to the form hook * Fix local form data when form isn't mounted If the form isn't on the page (e.g. if we're read-only), then useFormData will return no values. In these cases, we can simply fall back to the initialState values, as they'll either be: the default values on a new form, or: the current values on an active create/edit form. Updates the manual type of useFormData to reflect this "maybe" fact. * Allow user to navigate between invalid tabs on Edit Rule * Form hooks now _always_ return the form's data, regardless of validity * Edit Rule now: * persists invalid data * submits both the active form and the destination form on navigation. This is necessary to refresh validations on the destination form, since the form lib assumes a newly-mounted form is valid * simplifies "invalid tab" logic to be derived from our persisted data * Fix logical error If the rule is immutable, they can only edit actions. * Remove unneeded eslint exception Fixed by upstream #76471 * Make 21 the default risk score for a new rule Since the default severity is 'low,' these two defaults now coincide. * Remove duplicated type in favor of common one

sebelga added 5 commits August 24, 2020 18:59

Refactor use_field and use_form to correctly add field

b852854

[Security solution] Update RuleActionsField comp to not rely on the f…

18adb6d

…orm object from field I removed the form object form the field object to avoid unnecessary effect triggers. The form object should be accessed by context with the "useFormContext()" hook.

Fix pipeline processors

e378a2f

Update mappings editor to use memoize config and validation for fields

31d9e3b

Fix component integration tests

4f80819

sebelga added bug Fixes for quality problems that affect the customer experience release_note:skip Skip the PR/issue when compiling release notes Team:Kibana Management Dev Tools, Index Management, Upgrade Assistant, ILM, Ingest Node Pipelines, and more v7.10.0 v8.0.0 labels Aug 24, 2020

sebelga added 6 commits August 25, 2020 08:33

Fix TS issue

abbda2c

[Form lib] FromDataProvider don't render anything until fields have m…

5afe601

…ounted

[Security solution] Fix riskScore form fields

73c86a7

[Security solution] Fix security level form fields

bca378b

[Form lib] Update form defaultValue whenever prop changes

d0f02db

Merge remote-tracking branch 'upstream/master' into bug/form-lib-addf…

4f8e5a7

…ield-on-component-mount

sebelga commented Aug 25, 2020

View reviewed changes

sebelga requested review from spong and jloleysens August 25, 2020 15:24

sebelga marked this pull request as ready for review August 25, 2020 15:24

sebelga requested review from a team as code owners August 25, 2020 15:24

Fix i18n issue

9ccde27

elasticmachine and others added 4 commits August 25, 2020 11:06

Merge branch 'master' into bug/form-lib-addfield-on-component-mount

e3d5594

Fix jest test

fb8de5c

Merge branch 'bug/form-lib-addfield-on-component-mount' of github.com…

55593c1

…:sebelga/kibana into bug/form-lib-addfield-on-component-mount

Merge remote-tracking branch 'upstream/master' into bug/form-lib-addf…

d04098b

…ield-on-component-mount

elasticmachine and others added 5 commits August 26, 2020 03:43

Merge branch 'master' into bug/form-lib-addfield-on-component-mount

ef8aa16

RiskScore: don't convert empty string to number

bd693da

Merge branch 'bug/form-lib-addfield-on-component-mount' of github.com…

20e723d

…:sebelga/kibana into bug/form-lib-addfield-on-component-mount

Add test coverage to FormDataProvider to subscribe to updated form value

44b5572

jloleysens approved these changes Aug 27, 2020

View reviewed changes

Merge branch 'master' into bug/form-lib-addfield-on-component-mount

b0c271f

patrykkopycinski approved these changes Aug 27, 2020

View reviewed changes

sebelga merged commit ec15160 into elastic:master Aug 27, 2020

sebelga deleted the bug/form-lib-addfield-on-component-mount branch August 27, 2020 12:02

sebelga mentioned this pull request Aug 27, 2020

[7.x] [Form lib] Correctly add field to form on component mount (#75796) #76077

Merged

sebelga mentioned this pull request Sep 3, 2020

[es_ui_shared] Fix eslint exhaustive deps rule #76392

Merged

peluja1012 mentioned this pull request Oct 14, 2020

[SIEM] Updating an existing detection rule shows the risk and criticality of the last created rule #78572

Closed

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

[Form lib] Correctly add field to form on component mount #75796

[Form lib] Correctly add field to form on component mount #75796

sebelga commented Aug 24, 2020 •

edited

Loading

sebelga Aug 25, 2020

rylnd Aug 25, 2020

sebelga Aug 26, 2020

rylnd Aug 26, 2020 •

edited

Loading

elasticmachine commented Aug 25, 2020

sebelga commented Aug 25, 2020

sebelga commented Aug 26, 2020

jloleysens left a comment

jloleysens Aug 27, 2020

jloleysens Aug 27, 2020

sebelga Aug 27, 2020

sebelga commented Aug 27, 2020

sebelga commented Aug 27, 2020

patrykkopycinski left a comment

kibanamachine commented Aug 27, 2020

sebelga commented Aug 27, 2020


		const areAllFieldsValidated = formFieldsValidity.every(({ 1: isValidated }) => isValidated);

		// If not all the fiels have been validated, the validity of the form is unknown, thus still "undefined"

	// If not all the fiels have been validated, the validity of the form is unknown, thus still "undefined"
	// If not all the fields have been validated, the validity of the form is unknown, thus still "undefined"

[Form lib] Correctly add field to form on component mount #75796

[Form lib] Correctly add field to form on component mount #75796

Conversation

sebelga commented Aug 24, 2020 • edited Loading

Security solution forms

Lesson learned

sebelga Aug 25, 2020

Choose a reason for hiding this comment

rylnd Aug 25, 2020

Choose a reason for hiding this comment

sebelga Aug 26, 2020

Choose a reason for hiding this comment

rylnd Aug 26, 2020 • edited Loading

Choose a reason for hiding this comment

elasticmachine commented Aug 25, 2020

sebelga commented Aug 25, 2020

sebelga commented Aug 26, 2020

jloleysens left a comment

Choose a reason for hiding this comment

jloleysens Aug 27, 2020

Choose a reason for hiding this comment

jloleysens Aug 27, 2020

Choose a reason for hiding this comment

sebelga Aug 27, 2020

Choose a reason for hiding this comment

sebelga commented Aug 27, 2020

sebelga commented Aug 27, 2020

patrykkopycinski left a comment

Choose a reason for hiding this comment

kibanamachine commented Aug 27, 2020

💚 Build Succeeded

Build metrics

async chunks size

page load bundle size

History

sebelga commented Aug 27, 2020

sebelga commented Aug 24, 2020 •

edited

Loading

rylnd Aug 26, 2020 •

edited

Loading