Update base for Update on "[caffe2] Remove IValue include from operator.h"

peterbell10 · peterbell10 · commit 2bccddc80b71 · 2021-10-22T20:22:48.000+01:00
ivalue.h includes Tensor.h, so creating a compilation barrier between
operator.h and ivalue.h means non-exported caffe2 ops don't need to be
rebuilt as often when developing PyTorch.

[ghstack-poisoned]
diff --git a/caffe2/core/operator.cc b/caffe2/core/operator.cc
@@ -127,7 +127,7 @@ OperatorBase::OperatorBase(
       input_size_(compute_input_size_(newstyle_inputs_)) {
   input_tensors_.resize(input_size_);
 
-  output_tensors_.reserve(outputs_.size());
+  output_tensors_.reserve(outputs.size());
   for (auto i : c10::irange(outputs.size())) {
     output_tensors_.emplace_back(outputs.extract(i));
   }

Original file line number	Diff line number	Diff line change
`@@ -127,7 +127,7 @@ OperatorBase::OperatorBase(`
`127`	`127`	`input_size_(compute_input_size_(newstyle_inputs_)) {`
`128`	`128`	`input_tensors_.resize(input_size_);`
`129`	`129`
`130`		`- output_tensors_.reserve(outputs_.size());`
	`130`	`+ output_tensors_.reserve(outputs.size());`
`131`	`131`	`for (auto i : c10::irange(outputs.size())) {`
`132`	`132`	`output_tensors_.emplace_back(outputs.extract(i));`
`133`	`133`	`}`