I/O Customizaton Rules: add support for changing the input type. #17347
Conversation
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
pcanal
requested review from
jblomer and
dpiparo
and removed request for
dpiparo
Test Results 2 files 2 suites 8h 51m 9s ⏱️ Results for commit 7b3edc4. ♻️ This comment has been updated with latest results. |
Previously the call to RemoveStreamerInfo was working only if the slot was the the slot of the current version. (the actually removed StreamerInfo was hard-coded to be the current version
The data members are cached (because they are input of a rule) need to be read into the original type not the 'current in memory type' so that the rule can properly function if the type has changed. We should upgrade further for the in-memory type of the input of a rule to be the one specified in the rule independently of the onfile type and the final in-memory type.
In the "list of real data", the array members' name is "datamembername[arraysizes]" in most cases. GetRealData already has code to find the real data if * input has brackets and name has brackets * input has brackets and name does not have brackets * input has no brackets and name has no brackets This commit adds: * input has no brackats and name has brackets. This is required for simplicity of searching from the code generated for I/O customization rules
In TStreamerInfo::InsertArtificialElements search for the offset of the data member based on the already calculated 'real data' spelling to speedup the lookup.
Previous, the inner code was looking an inexistent members (actually the name of the object in its parent rather than an inner data member) and was appearing to work because the offset return in case of error was '0' (which worked since indeed there was no offset to be added to the address we already calculated)
Instead of returning 0 when not finding any information about the data member, `TClass::GetDataMemberOffset` now returns TVirtualStreamerInfo::kMissing. This allows the reading of the element to be "skipped" rather than having it over-write other members
Return the underlying type and other piece and information separately
This reverts commit e900d32. Bug fix: I/O customization rules which missing input are no longer (inappropriately) run. We now properly warn if a rule is requesting an input that is not present in the StreamerInfo that the rule is being applied to. To avoid the warning in some cases (in particular the cases used to support `HepMC` and are located in `$ROOTSYS/etc/class.rules`) we introduce a new "attribute" for I/O customization rules: `CanIgnore:` When using this attribute the rule will be ignored if the input is missing from the schema/class-layout they apply to instead of issue a `Warning` wip: Revert "Don't warn when ..."
In TStreamerInfo's Build and BuildOld update the artificial StreamerElement's description of the in memory type to be used to store the data. This allows the rule to request an implicit transformation from onfile representation to in memory representation before running the explicit customization rules
This allows to remove data member whose type is array of (sub)objects.
Previously the code in TStreamerElement assumed that GetClassPointer return the underlying type of the in memory representation. This was true until recently as you could not change the representation in the 'current' StreamerInfo ( GetClassPointer was matching the layout as decribed by the Clang AST ) and if the class was not loaded, the 'old/onfile' representation was the only possible choice. However we now can change the representation on file (i.e. GetClassPointer) for the current StreamerInfo of a loaded class (i.e. later using Write rule, now in the case of enum class and collection thereof) and in the case of the synthetic/artifical class created to cache inputs of rules (they are named `classname@@versionNumber`), the rule now dictates the representation in memory
pcanal
force-pushed
the
io-rules-type-change
branch
from
ba28c30 to
7b3edc4
Compare
jblomer
reviewed
Jan 8, 2025
| @@ -7403,7 +7403,7 @@ void TClass::RemoveStreamerInfo(Int_t slot) | |||
| if (fStreamerInfo->GetSize() >= slot) { | |||
| R__LOCKGUARD(gInterpreterMutex); | |||
| TVirtualStreamerInfo *info = (TVirtualStreamerInfo*)fStreamerInfo->At(slot); | |||
| fStreamerInfo->RemoveAt(fClassVersion); | |||
| fStreamerInfo->RemoveAt(slot); | |||
There was a problem hiding this comment.
Can we add a test for adding and removing streamer infos?
| } | ||
| if (objName == nameNoDim) { | ||
| return static_cast<TRealData*>(obj); |
There was a problem hiding this comment.
Can we add a test for the 4 cases?
Comment on lines
+548
to
+550
| Error("Setup","%s",Form("Negative offset %d for %s in %s, class: %s", | ||
| fMemberOffset, fDataMember.Data(), fBranch->GetName(), fClass->GetName())); | ||
| fMemberOffset = 0; |
There was a problem hiding this comment.
Can this happen? If so, can we add a test for this?
| } | ||
|
|
||
| } else if (fClass) { | ||
| } else if (fParent && fClass) { |
There was a problem hiding this comment.
Which case do we cover by the additional check for fParent?
io/io/src/TStreamerInfo.cxx
Outdated
| @@ -253,6 +253,62 @@ namespace { | |||
| TStreamerInfo* fInfo; | |||
| }; | |||
|
|
|||
| decltype(auto) GetSourceType(ROOT::TSchemaRule::TSources *s) | |||
There was a problem hiding this comment.
Why do we need decltype(auto) instead of the spelled out type std::tuple<...>?
Comment on lines
+333
to
+335
| for (Int_t j=0;j<compinfo->fLength;j++) { | ||
| b.SkipObjectAny(); | ||
| } |
There was a problem hiding this comment.
I think this is not tested by root-project/roottest#1247. In this case, can a test be added?
| @@ -257,13 +257,14 @@ namespace { | |||
| { | |||
| std::string localtypename(s->GetUnderlyingTypeName()); | |||
| Int_t ndim = 0; | |||
| Int_t totaldim = 1; | |||
| Int_t totaldim = 0; | |||
There was a problem hiding this comment.
Perhaps merge this with the commit adding totaldim
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
With these changes, one can request the schema evolution (usually implicit) of the input data members from the on-file type to a custom type. Example of the changes support in the test roottest/root/io/evolution/rules/execSourceTypes.cxx added by the companion roottest PR
This fixes #17346 17346