Close https://github.com/llvm/llvm-project/issues/90154
This patch is also an optimization to the lookup process to utilize the
information provided by `export` keyword.
Previously, in the lookup process, the `export` keyword only takes part
in the check part, it doesn't get involved in the lookup process. That
said, previously, in a name lookup for 'name', we would load all of
declarations with the name 'name' and check if these declarations are
valid or not. It works well. But it is inefficient since it may load
declarations that may not be wanted.
Note that this patch actually did a trick in the lookup process instead
of bring module information to DeclarationName or considering module
information when deciding if two declarations are the same. So it may
not be a surprise to me if there are missing cases. But it is not a
regression. It should be already the case. Issue reports are welcomed.
In this patch, I tried to split the big lookup table into a lookup table
as before and a module local lookup table, which takes a combination of
the ID of the DeclContext and hash value of the primary module name as
the key. And refactored `DeclContext::lookup()` method to take the
module information. So that a lookup in a DeclContext won't load
declarations that are local to **other** modules.
And also I think it is already beneficial to split the big lookup table
since it may reduce the conflicts during lookups in the hash table.
BTW, this patch introduced a **regression** for a reachability rule in
C++20 but it was false-negative. See
'clang/test/CXX/module/module.interface/p7.cpp' for details.
This patch is not expected to introduce any other
regressions for non-c++20-modules users since the module local lookup
table should be empty for them.
Close https://github.com/llvm/llvm-project/issues/90154
This patch is also an optimization to the lookup process to utilize the
information provided by `export` keyword.
Previously, in the lookup process, the `export` keyword only takes part
in the check part, it doesn't get involved in the lookup process. That
said, previously, in a name lookup for 'name', we would load all of
declarations with the name 'name' and check if these declarations are
valid or not. It works well. But it is inefficient since it may load
declarations that may not be wanted.
Note that this patch actually did a trick in the lookup process instead
of bring module information to DeclarationName or considering module
information when deciding if two declarations are the same. So it may
not be a surprise to me if there are missing cases. But it is not a
regression. It should be already the case. Issue reports are welcomed.
In this patch, I tried to split the big lookup table into a lookup table
as before and a module local lookup table, which takes a combination of
the ID of the DeclContext and hash value of the primary module name as
the key. And refactored `DeclContext::lookup()` method to take the
module information. So that a lookup in a DeclContext won't load
declarations that are local to **other** modules.
And also I think it is already beneficial to split the big lookup table
since it may reduce the conflicts during lookups in the hash table.
BTW, this patch introduced a **regression** for a reachability rule in
C++20 but it was false-negative. See
'clang/test/CXX/module/module.interface/p7.cpp' for details.
This patch is not expected to introduce any other
regressions for non-c++20-modules users since the module local lookup
table should be empty for them.
---
On the API side, this patch unfortunately add a maybe-confusing argument
`Module *NamedModule` to
`ExternalASTSource::FindExternalVisibleDeclsByName()`. People may think
we can get the information from the first argument `const DeclContext
*DC`. But sadly there are declarations (e.g., namespace) can appear in
multiple different modules as a single declaration. So we have to add
additional information to indicate this.
This is a new Clang-specific attribute to ensure that field
initializations are performed explicitly.
For example, if we have
```
struct B {
[[clang::explicit]] int f1;
};
```
then the diagnostic would trigger if we do `B b{};`:
```
field 'f1' is left uninitialized, but was marked as requiring initialization
```
This prevents callers from accidentally forgetting to initialize fields,
particularly when new fields are added to the class.
After 0dedd6fe1 and 03229e7c0, invalid concept declarations might lack
expressions for evaluation and normalization. This could make it crash
in certain scenarios, apart from the one of evaluation concepts showed
in 03229e7c0, there's also an issue when checking specializations where
the normalization also relies on a non-null expression.
This patch prevents that by avoiding building up a type constraint in
such situations, thereafter the template parameter wouldn't have a
concept specialization of a null expression.
With this patch, the assumption in ASTWriterDecl is no longer valid.
Namely, HasConstraint and TypeConstraintInitialized must now represent
different meanings for both source fidelity and semantic requirements.
Fixes https://github.com/llvm/llvm-project/issues/115004
Fixes https://github.com/llvm/llvm-project/issues/121980
Reland https://github.com/llvm/llvm-project/pull/83237
---
(Original comments)
Currently all the specializations of a template (including
instantiation, specialization and partial specializations) will be
loaded at once if we want to instantiate another instance for the
template, or find instantiation for the template, or just want to
complete the redecl chain.
This means basically we need to load every specializations for the
template once the template declaration got loaded. This is bad since
when we load a specialization, we need to load all of its template
arguments. Then we have to deserialize a lot of unnecessary
declarations.
For example,
```
// M.cppm
export module M;
export template <class T>
class A {};
export class ShouldNotBeLoaded {};
export class Temp {
A<ShouldNotBeLoaded> AS;
};
// use.cpp
import M;
A<int> a;
```
We have a specialization ` A<ShouldNotBeLoaded>` in `M.cppm` and we
instantiate the template `A` in `use.cpp`. Then we will deserialize
`ShouldNotBeLoaded` surprisingly when compiling `use.cpp`. And this
patch tries to avoid that.
Given that the templates are heavily used in C++, this is a pain point
for the performance.
This patch adds MultiOnDiskHashTable for specializations in the
ASTReader. Then we will only deserialize the specializations with the
same template arguments. We made that by using ODRHash for the template
arguments as the key of the hash table.
To review this patch, I think `ASTReaderDecl::AddLazySpecializations`
may be a good entry point.
Note that PointerUnion::{is,get} have been soft deprecated in
PointerUnion.h:
// FIXME: Replace the uses of is(), get() and dyn_cast() with
// isa<T>, cast<T> and the llvm::dyn_cast<T>
I'm not touching PointerUnion::dyn_cast for now because it's a bit
complicated; we could blindly migrate it to dyn_cast_if_present, but
we should probably use dyn_cast when the operand is known to be
non-null.
Currently all the specializations of a template (including
instantiation, specialization and partial specializations) will be
loaded at once if we want to instantiate another instance for the
template, or find instantiation for the template, or just want to
complete the redecl chain.
This means basically we need to load every specializations for the
template once the template declaration got loaded. This is bad since
when we load a specialization, we need to load all of its template
arguments. Then we have to deserialize a lot of unnecessary
declarations.
For example,
```
// M.cppm
export module M;
export template <class T>
class A {};
export class ShouldNotBeLoaded {};
export class Temp {
A<ShouldNotBeLoaded> AS;
};
// use.cpp
import M;
A<int> a;
```
We should a specialization ` A<ShouldNotBeLoaded>` in `M.cppm` and we
instantiate the template `A` in `use.cpp`. Then we will deserialize
`ShouldNotBeLoaded` surprisingly when compiling `use.cpp`. And this
patch tries to avoid that.
Given that the templates are heavily used in C++, this is a pain point
for the performance.
This patch adds MultiOnDiskHashTable for specializations in the
ASTReader. Then we will only deserialize the specializations with the
same template arguments. We made that by using ODRHash for the template
arguments as the key of the hash table.
To review this patch, I think `ASTReaderDecl::AddLazySpecializations`
may be a good entry point.
The patch was reviewed in
https://github.com/llvm/llvm-project/pull/83237 but that PR is a stacked
PR. But I feel the intention of the stacked PRs get lost during the
review process. So I feel it is better to merge the commits into a
single commit instead of merging them in the PR page. It is better for
us to cherry-pick and revert.
This patch removes `ASTWriter::Context` and starts passing `ASTContext
&` explicitly to functions that actually need it. This is a
non-functional change with the end-goal of being able to write
lightweight PCM files with no `ASTContext` at all.
We made the incorrect assumption that names of fields are unique when
creating their default initializers.
We fix that by keeping track of the instantiaation pattern for field
decls that are placeholder vars,
like we already do for unamed fields.
Fixes#114069
This patch reapplies #111173, fixing a bug when instantiating dependent
expressions that name a member template that is later explicitly
specialized for a class specialization that is implicitly instantiated.
The bug is addressed by adding the `hasMemberSpecialization` function,
which return `true` if _any_ redeclaration is a member specialization.
This is then used when determining the instantiation pattern for a
specialization of a template, and when collecting template arguments for
a specialization of a template.
This fixes instantiation of definition for friend function templates,
when the declaration found and the one containing the definition
have different template contexts.
In these cases, the the function declaration corresponding to the
definition is not available; it may not even be instantiated at all.
So this patch adds a bit which tracks which function template
declaration was instantiated from the member template.
It's used to find which primary template serves as a context
for the purpose of obtaining the template arguments needed
to instantiate the definition.
Fixes#55509
Reapplies #106585, fixing an issue where non-dependent names of member
templates appearing prior to that member template being explicitly
specialized for an implicitly instantiated class template specialization
would incorrectly use the definition of the explicitly specialized
member template.
Fixed a crash for the attached test case due to we missed to emit the
deduction guide. The reason is, the deduction guide is attached to the
export-decl in the imported module. So we won't emit it by traversing the
AST of the current TU.
Summary:
https://github.com/llvm/llvm-project/pull/109167 serializes
FunctionToLambdasMap in the order of pointers in DenseMap. It gives
different order with different memory layouts. Fix this issue by using
LocalDeclID instead of pointers.
Test Plan: check-clang
Summary:
Because AST loading code is lazy and happens in unpredictable order, it
is possible that a function and lambda inside the function can be loaded
from different modules. As a result, the captured DeclRefExpr won’t
match the corresponding VarDecl inside the function. This situation is
reflected in the AST as follows:
```
FunctionDecl 0x555564f4aff0 <Conv.h:33:1, line:41:1> line:33:35 imported in ./thrift_cpp2_base.h hidden tryTo 'Expected<Tgt, const char *> ()' inline
|-also in ./folly-conv.h
`-CompoundStmt 0x555564f7cfc8 <col:43, line:41:1>
|-DeclStmt 0x555564f7ced8 <line:34:3, col:17>
| `-VarDecl 0x555564f7cef8 <col:3, col:16> col:7 imported in ./thrift_cpp2_base.h hidden referenced result 'Tgt' cinit
| `-IntegerLiteral 0x555564f7d080 <col:16> 'int' 0
|-CallExpr 0x555564f7cea8 <line:39:3, col:76> '<dependent type>'
| |-UnresolvedLookupExpr 0x555564f7bea0 <col:3, col:19> '<overloaded function type>' lvalue (no ADL) = 'then_' 0x555564f7bef0
| |-CXXTemporaryObjectExpr 0x555564f7bcb0 <col:25, col:45> 'Expected<bool, int>':'folly::Expected<bool, int>' 'void () noexcept' zeroing
| `-LambdaExpr 0x555564f7bc88 <col:48, col:75> '(lambda at Conv.h:39:48)'
| |-CXXRecordDecl 0x555564f76b88 <col:48> col:48 imported in ./folly-conv.h hidden implicit <undeserialized declarations> class definition
| | |-also in ./thrift_cpp2_base.h
| | `-DefinitionData lambda empty standard_layout trivially_copyable literal can_const_default_init
| | |-DefaultConstructor defaulted_is_constexpr
| | |-CopyConstructor simple trivial has_const_param needs_implicit implicit_has_const_param
| | |-MoveConstructor exists simple trivial needs_implicit
| | |-CopyAssignment trivial has_const_param needs_implicit implicit_has_const_param
| | |-MoveAssignment
| | `-Destructor simple irrelevant trivial constexpr needs_implicit
| `-CompoundStmt 0x555564f7d1a8 <col:58, col:75>
| `-ReturnStmt 0x555564f7d198 <col:60, col:67>
| `-DeclRefExpr 0x555564f7d0a0 <col:67> 'Tgt' lvalue Var 0x555564f7d0c8 'result' 'Tgt' refers_to_enclosing_variable_or_capture
`-ReturnStmt 0x555564f7bc78 <line:40:3, col:11>
`-InitListExpr 0x555564f7bc38 <col:10, col:11> 'void'
```
This diff modifies the AST deserialization process to load lambdas
within the canonical function declaration sooner, immediately following
the function, ensuring that they are loaded from the same module.
Re-land https://github.com/llvm/llvm-project/pull/104512 Added test case
that caused crash due to multiple enclosed lambdas deserialization.
Test Plan: check-clang
Currently, clang rejects the following explicit specialization of `f`
due to the constraints not being equivalent:
```
template<typename T>
struct A
{
template<bool B>
void f() requires B;
};
template<>
template<bool B>
void A<int>::f() requires B { }
```
This happens because, in most cases, we do not set the flag indicating
whether a `RedeclarableTemplate` is an explicit specialization of a
member of an implicitly instantiated class template specialization until
_after_ we compare constraints for equivalence. This patch addresses the
issue (and a number of other issues) by:
- storing the flag indicating whether a declaration is a member
specialization on a per declaration basis, and
- significantly refactoring `Sema::getTemplateInstantiationArgs` so we
collect the right set of template argument in all cases.
Many of our declaration matching & constraint evaluation woes can be
traced back to bugs in `Sema::getTemplateInstantiationArgs`. This
change/refactor should fix a lot of them. It also paves the way for
fixing #101330 and #105462 per my suggestion in #102267 (which I have
implemented on top of this patch but will merge in a subsequent PR).
Reland https://github.com/llvm/llvm-project/pull/75912
The differences of this PR between
https://github.com/llvm/llvm-project/pull/75912 are:
- Fixed a regression in `Decl::isInAnotherModuleUnit()` in DeclBase.cpp
pointed by @mizvekov and add the corresponding test.
- Fixed the regression in windows
https://github.com/llvm/llvm-project/issues/97447. The changes are in
`CodeGenModule::getVTableLinkage` from
`clang/lib/CodeGen/CGVTables.cpp`. According to the feedbacks from MSVC
devs, the linkage of vtables won't affected by modules. So I simply
skipped the case for MSVC.
Given this is more or less fundamental to the use of modules. I hope we
can backport this to 19.x.
Previously, we skipped calculating ODRHash for decls in GMF when writing
them to .pcm files as an optimization. But actually, it is not
true that this will be a pure optimization. Whether or not it is
beneficial depends on the use cases. For example, if we're writing a
function `a` in module and there are 10 consumers of `a` in other TUs,
then the other TUs will pay for the cost to calculate the ODR hash for
`a` ten times. Then this optimization doesn't work. However, if all the
consumers of the module didn't touch `a`, then we can save the cost to
calculate the ODR hash of `a` for 1 times.
And the assumption to make it was: generally, the consumers of a module
may only consume a small part of the imported module. This is the reason
why we tried to load declarations, types and identifiers lazily. Then it
looks good to do the similar thing for calculating ODR hashs.
It works fine for a long time, until we started to look into the support
of modules in clangd. Then we meet multiple issue reports complaining
we're calculating ODR hash in the wrong place. To workaround these issue
reports, I decided to always write the ODRhash for decls in GMF. In my
local test, I only observed less than 1% compile time regression after
doing this. So it should be fine.
Currently, `NamespaceDecl` has a member `AnonOrFirstNamespaceAndFlags`
which stores a few pieces of data:
- a bit indicating whether the namespace was declared `inline`, and
- a bit indicating whether the namespace was declared as a
_nested-namespace-definition_, and
- a pointer a `NamespaceDecl` that either stores:
- a pointer to the first declaration of that namespace if the
declaration is no the first declaration, or
- a pointer to the unnamed namespace that inhabits the namespace
otherwise.
`Redeclarable` already stores a pointer to the first declaration of an
entity, so it's unnecessary to store this in `NamespaceDecl`.
`DeclContext` has 8 bytes in which various bitfields can be stored for a
declaration, so it's not necessary to store these in `NamespaceDecl`
either. We only need to store a pointer to the unnamed namespace that
inhabits the first declaration of a namespace. This patch moves the two
bits currently stored in `NamespaceDecl` to `DeclContext`, and only
stores a pointer to the unnamed namespace that inhabits a namespace in
the first declaration of that namespace. Since `getOriginalNamespace`
always returns the same `NamespaceDecl` as `getFirstDecl`, this function
is removed to avoid confusion.
This reverts commit 18f3bcbb13ca83d33223b00761d8cddf463e9ffb, 15bb02650e26875c48889053d6a9697444583721 and
99873b35da7ecb905143c8a6b8deca4d4416f1a9.
See the post commit message in
https://github.com/llvm/llvm-project/pull/75912 to see the reasons.
Now we can create a LocalDeclID directly with an integer without
verifying. It may be hard to refactor if we want to change the way we
serialize DeclIDs (See https://github.com/llvm/llvm-project/pull/95897).
Also it is hard for us to debug if someday someone construct a
LocalDeclID with an incorrect value.
So in this patch, I tried to unify the way we can construct a
LocalDeclID in ASTReader, where we will construct the LocalDeclID from
the serialized data. Also, now we can verify the constructed LocalDeclID
sooner in the new interface.
carefully
Close https://github.com/llvm/llvm-project/issues/93859
The direct pattern of the issue is that, in a reduced BMI, we're going
to wrtie a class but we didn't write the deduction guide. Although we
handled deduction guide, but we tried to record the found deduction
guide from `noload_lookup` directly.
It is slightly problematic if the found deduction guide is from AST.
e.g.,
```
module;
export module m;
import xxx; // Also contains the class and the deduction guide
...
```
Then when we writes the class in the current file, we tried to record
the deduction guide, but `noload_lookup` returns the deduction guide
from the AST file then we didn't record the local deduction guide. Then
mismatch happens.
To mitiagte the problem, we tried to record the canonical declaration
for the decution guide.
This is an enabler for https://github.com/llvm/llvm-project/pull/92855
This allows an NTTP default argument to be set as an arbitrary
TemplateArgument, not just an expression.
This allows template parameter packs to have default arguments in the
AST, even though the language proper doesn't support the syntax for it.
This allows NTTP default arguments to be other kinds of arguments, like
packs, integral constants, and such.
This is an enabler for a future patch.
This allows an type-parameter default argument to be set as an arbitrary
TemplateArgument, not just a type.
This allows template parameter packs to have default arguments in the
AST, even though the language proper doesn't support the syntax for it.
This will be used in a later patch which synthesizes template parameter
lists with arbitrary default arguments taken from template
specializations.
There are a few places we used SubsType, because we only had a type, now
we use SubstTemplateArgument.
SubstTemplateArgument was missing arguments for setting Instantiation
location and entity names.
Adding those is needed so we don't regress in diagnostics.
Our current method of storing the template arguments as written for
`(Class/Var)Template(Partial)SpecializationDecl` suffers from a number
of flaws:
- We use `TypeSourceInfo` to store `TemplateArgumentLocs` for class
template/variable template partial/explicit specializations. For
variable template specializations, this is a rather unintuitive hack (as
we store a non-type specialization as a type). Moreover, we don't ever
*need* the type as written -- in almost all cases, we only want the
template arguments (e.g. in tooling use-cases).
- The template arguments as written are stored in a number of redundant
data members. For example, `(Class/Var)TemplatePartialSpecialization`
have their own `ArgsAsWritten` member that stores an
`ASTTemplateArgumentListInfo` (the template arguments).
`VarTemplateSpecializationDecl` has yet _another_ redundant member
"`TemplateArgsInfo`" that also stores an `ASTTemplateArgumentListInfo`.
This patch eliminates all
`(Class/Var)Template(Partial)SpecializationDecl` members which store the
template arguments as written, and turns the `ExplicitInfo` member into
a `llvm::PointerUnion<const ASTTemplateArgumentListInfo*,
ExplicitInstantiationInfo*>` (to avoid unnecessary allocations when the
declaration isn't an explicit instantiation). The template arguments as
written are now accessed via `getTemplateArgsWritten` in all cases.
The "most breaking" change is to AST Matchers, insofar that `hasTypeLoc`
will no longer match class template specializations (since they no
longer store the type as written).
This relands 6c31104.
The patch was reverted due to incorrectly introduced alignment. And the
patch was re-commited after fixing the alignment issue.
Following off are the original message:
This is part of "no transitive change" patch series, "no transitive
source location change". I talked this with @Bigcheese in the tokyo's
WG21 meeting.
The idea comes from @jyknight posted on LLVM discourse. That for:
```
// A.cppm
export module A;
...
// B.cppm
export module B;
import A;
...
//--- C.cppm
export module C;
import C;
```
Almost every time A.cppm changes, we need to recompile `B`. Due to we
think the source location is significant to the semantics. But it may be
good if we can avoid recompiling `C` if the change from `A` wouldn't
change the BMI of B.
This patch only cares source locations. So let's focus on source
location's example. We can see the full example from the attached test.
```
//--- A.cppm
export module A;
export template <class T>
struct C {
T func() {
return T(43);
}
};
export int funcA() {
return 43;
}
//--- A.v1.cppm
export module A;
export template <class T>
struct C {
T func() {
return T(43);
}
};
export int funcA() {
return 43;
}
//--- B.cppm
export module B;
import A;
export int funcB() {
return funcA();
}
//--- C.cppm
export module C;
import A;
export void testD() {
C<int> c;
c.func();
}
```
Here the only difference between `A.cppm` and `A.v1.cppm` is that
`A.v1.cppm` has an additional blank line. Then the test shows that two
BMI of `B.cppm`, one specified `-fmodule-file=A=A.pcm` and the other
specified `-fmodule-file=A=A.v1.pcm`, should have the bit-wise same
contents.
However, it is a different story for C, since C instantiates templates
from A, and the instantiation records the source information from module
A, which is different from `A` and `A.v1`, so it is expected that the
BMI `C.pcm` and `C.v1.pcm` can and should differ.
To fully understand the patch, we need to understand how we encodes
source locations and how we serialize and deserialize them.
For source locations, we encoded them as:
```
|
|
| _____ base offset of an imported module
|
|
|
|_____ base offset of another imported module
|
|
|
|
| ___ 0
```
As the diagram shows, we encode the local (unloaded) source location
from 0 to higher bits. And we allocate the space for source locations
from the loaded modules from high bits to 0. Then the source locations
from the loaded modules will be mapped to our source location space
according to the allocated offset.
For example, for,
```
// a.cppm
export module a;
...
// b.cppm
export module b;
import a;
...
```
Assuming the offset of a source location (let's name the location as
`S`) in a.cppm is 45 and we will record the value `45` into the BMI
`a.pcm`. Then in b.cppm, when we import a, the source manager will
allocate a space for module 'a' (according to the recorded number of
source locations) as the base offset of module 'a' in the current source
location spaces. Let's assume the allocated base offset as 90 in this
example. Then when we want to get the location in the current source
location space for `S`, we can get it simply by adding `45` to `90` to
`135`. Finally we can get the source location for `S` in module B as
`135`.
And when we want to write module `b`, we would also write the source
location of `S` as `135` directly in the BMI. And to clarify the
location `S` comes from module `a`, we also need to record the base
offset of module `a`, 90 in the BMI of `b`.
Then the problem comes. Since the base offset of module 'a' is computed
by the number source locations in module 'a'. In module 'b', the
recorded base offset of module 'a' will change every time the number of
source locations in module 'a' increase or decrease. In other words, the
contents of BMI of B will change every time the number of locations in
module 'a' changes. This is pretty sensitive. Almost every change will
change the number of locations. So this is the problem this patch want
to solve.
Let's continue with the existing design to understand what's going on.
Another interesting case is:
```
// c.cppm
export module c;
import whatever;
import a;
import b;
...
```
In `c.cppm`, when we import `a`, we still need to allocate a base
location offset for it, let's say the value becomes to `200` somehow.
Then when we reach the location `S` recorded in module `b`, we need to
translate it into the current source location space. The solution is
quite simple, we can get it by `135 + (200 - 90) = 245`. In another
word, the offset of a source location in current module can be computed
as `Recorded Offset + Base Offset of the its module file - Recorded Base
Offset`.
Then we're almost done about how we handle the offset of source
locations in serializers.
From the abstract level, what we want to do is to remove the hardcoded
base offset of imported modules and remain the ability to calculate the
source location in a new module unit. To achieve this, we need to be
able to find the module file owning a source location from the encoding
of the source location.
So in this patch, for each source location, we will store the local
offset of the location and the module file index. For the above example,
in `b.pcm`, the source location of `S` will be recorded as `135`
directly. And in the new design, the source location of `S` will be
recorded as `<1, 45>`. Here `1` stands for the module file index of `a`
in module `b`. And `45` means the offset of `S` to the base offset of
module `a`.
So the trade-off here is that, to make the BMI more independent, we need
to record more abstract information. And I feel it is worthy. The
recompilation problem of modules is really annoying and there are still
people complaining this. But if we can make this (including stopping
other changes transitively), I think this may be a killer feature for
modules. And from @Bigcheese , this should be helpful for clang explicit
modules too.
And the benchmarking side, I tested this patch against
https://github.com/alibaba/async_simple/tree/CXX20Modules. No
significant change on compilation time. The size of .pcm files becomes
to 204M from 200M. I think the trade-off is pretty fair.
I didn't use another slot to record the module file index. I tried to
use the higher 32 bits of the existing source location encodings to
store that information. This design may be safe. Since we use `unsigned`
to store source locations but we use uint64_t in serialization. And
generally `unsigned` is 32 bit width in most platforms. So it might not
be a safe problem. Since all the bits we used to store the module file
index is not used before. So the new encodings may be:
```
|-----------------------|-----------------------|
| A | B | C |
* A: 32 bit. The index of the module file in the module manager + 1.
* The +1
here is necessary since we wish 0 stands for the current
module file.
* B: 31 bit. The offset of the source location to the module file
* containing it.
* C: The macro bit. We rotate it to the lowest bit so that we can save
* some
space in case the index of the module file is 0.
```
(The B and C is the existing raw encoding for source locations)
Another reason to reuse the same slot of the source location is to
reduce the impact of the patch. Since there are a lot of places assuming
we can store and get a source location from a slot. And if I tried to
add another slot, a lot of codes breaks. I don't feel it is worhty.
Another impact of this decision is that, the existing small
optimizations for encoding source location may be invalided. The key of
the optimization is that we can turn large values into small values then
we can use VBR6 format to reduce the size. But if we decided to put the
module file index into the higher bits, then maybe it simply doesn't
work. An example may be the `SourceLocationSequence` optimization.
This will only affect the size of on-disk .pcm files. I don't expect
this impact the speed and memory use of compilations. And seeing my
small experiments above, I feel this trade off is worthy.
The mental model for handling source location offsets is not so complex
and I believe we can solve it by adding module file index to each stored
source location.
For the practical side, since the source location is pretty sensitive,
and the patch can pass all the in-tree tests and a small scale projects,
I feel it should be correct.
I'll continue to work on no transitive decl change and no transitive
identifier change (if matters) to achieve the goal to stop the
propagation of unnecessary changes. But all of this depends on this
patch. Since, clearly, the source locations are the most sensitive
thing.
---
The release nots and documentation will be added seperately.
This is part of "no transitive change" patch series, "no transitive
source location change". I talked this with @Bigcheese in the tokyo's
WG21 meeting.
The idea comes from @jyknight posted on LLVM discourse. That for:
```
// A.cppm
export module A;
...
// B.cppm
export module B;
import A;
...
//--- C.cppm
export module C;
import C;
```
Almost every time A.cppm changes, we need to recompile `B`. Due to we
think the source location is significant to the semantics. But it may be
good if we can avoid recompiling `C` if the change from `A` wouldn't
change the BMI of B.
# Motivation Example
This patch only cares source locations. So let's focus on source
location's example. We can see the full example from the attached test.
```
//--- A.cppm
export module A;
export template <class T>
struct C {
T func() {
return T(43);
}
};
export int funcA() {
return 43;
}
//--- A.v1.cppm
export module A;
export template <class T>
struct C {
T func() {
return T(43);
}
};
export int funcA() {
return 43;
}
//--- B.cppm
export module B;
import A;
export int funcB() {
return funcA();
}
//--- C.cppm
export module C;
import A;
export void testD() {
C<int> c;
c.func();
}
```
Here the only difference between `A.cppm` and `A.v1.cppm` is that
`A.v1.cppm` has an additional blank line. Then the test shows that two
BMI of `B.cppm`, one specified `-fmodule-file=A=A.pcm` and the other
specified `-fmodule-file=A=A.v1.pcm`, should have the bit-wise same
contents.
However, it is a different story for C, since C instantiates templates
from A, and the instantiation records the source information from module
A, which is different from `A` and `A.v1`, so it is expected that the
BMI `C.pcm` and `C.v1.pcm` can and should differ.
# Internal perspective of status quo
To fully understand the patch, we need to understand how we encodes
source locations and how we serialize and deserialize them.
For source locations, we encoded them as:
```
|
|
| _____ base offset of an imported module
|
|
|
|_____ base offset of another imported module
|
|
|
|
| ___ 0
```
As the diagram shows, we encode the local (unloaded) source location
from 0 to higher bits. And we allocate the space for source locations
from the loaded modules from high bits to 0. Then the source locations
from the loaded modules will be mapped to our source location space
according to the allocated offset.
For example, for,
```
// a.cppm
export module a;
...
// b.cppm
export module b;
import a;
...
```
Assuming the offset of a source location (let's name the location as
`S`) in a.cppm is 45 and we will record the value `45` into the BMI
`a.pcm`. Then in b.cppm, when we import a, the source manager will
allocate a space for module 'a' (according to the recorded number of
source locations) as the base offset of module 'a' in the current source
location spaces. Let's assume the allocated base offset as 90 in this
example. Then when we want to get the location in the current source
location space for `S`, we can get it simply by adding `45` to `90` to
`135`. Finally we can get the source location for `S` in module B as
`135`.
And when we want to write module `b`, we would also write the source
location of `S` as `135` directly in the BMI. And to clarify the
location `S` comes from module `a`, we also need to record the base
offset of module `a`, 90 in the BMI of `b`.
Then the problem comes. Since the base offset of module 'a' is computed
by the number source locations in module 'a'. In module 'b', the
recorded base offset of module 'a' will change every time the number of
source locations in module 'a' increase or decrease. In other words, the
contents of BMI of B will change every time the number of locations in
module 'a' changes. This is pretty sensitive. Almost every change will
change the number of locations. So this is the problem this patch want
to solve.
Let's continue with the existing design to understand what's going on.
Another interesting case is:
```
// c.cppm
export module c;
import whatever;
import a;
import b;
...
```
In `c.cppm`, when we import `a`, we still need to allocate a base
location offset for it, let's say the value becomes to `200` somehow.
Then when we reach the location `S` recorded in module `b`, we need to
translate it into the current source location space. The solution is
quite simple, we can get it by `135 + (200 - 90) = 245`. In another
word, the offset of a source location in current module can be computed
as `Recorded Offset + Base Offset of the its module file - Recorded Base
Offset`.
Then we're almost done about how we handle the offset of source
locations in serializers.
# The high level design of current patch
From the abstract level, what we want to do is to remove the hardcoded
base offset of imported modules and remain the ability to calculate the
source location in a new module unit. To achieve this, we need to be
able to find the module file owning a source location from the encoding
of the source location.
So in this patch, for each source location, we will store the local
offset of the location and the module file index. For the above example,
in `b.pcm`, the source location of `S` will be recorded as `135`
directly. And in the new design, the source location of `S` will be
recorded as `<1, 45>`. Here `1` stands for the module file index of `a`
in module `b`. And `45` means the offset of `S` to the base offset of
module `a`.
So the trade-off here is that, to make the BMI more independent, we need
to record more abstract information. And I feel it is worthy. The
recompilation problem of modules is really annoying and there are still
people complaining this. But if we can make this (including stopping
other changes transitively), I think this may be a killer feature for
modules. And from @Bigcheese , this should be helpful for clang explicit
modules too.
And the benchmarking side, I tested this patch against
https://github.com/alibaba/async_simple/tree/CXX20Modules. No
significant change on compilation time. The size of .pcm files becomes
to 204M from 200M. I think the trade-off is pretty fair.
# Some low level details
I didn't use another slot to record the module file index. I tried to
use the higher 32 bits of the existing source location encodings to
store that information. This design may be safe. Since we use `unsigned`
to store source locations but we use uint64_t in serialization. And
generally `unsigned` is 32 bit width in most platforms. So it might not
be a safe problem. Since all the bits we used to store the module file
index is not used before. So the new encodings may be:
```
|-----------------------|-----------------------|
| A | B | C |
* A: 32 bit. The index of the module file in the module manager + 1. The +1
here is necessary since we wish 0 stands for the current module file.
* B: 31 bit. The offset of the source location to the module file containing it.
* C: The macro bit. We rotate it to the lowest bit so that we can save some
space in case the index of the module file is 0.
```
(The B and C is the existing raw encoding for source locations)
Another reason to reuse the same slot of the source location is to
reduce the impact of the patch. Since there are a lot of places assuming
we can store and get a source location from a slot. And if I tried to
add another slot, a lot of codes breaks. I don't feel it is worhty.
Another impact of this decision is that, the existing small
optimizations for encoding source location may be invalided. The key of
the optimization is that we can turn large values into small values then
we can use VBR6 format to reduce the size. But if we decided to put the
module file index into the higher bits, then maybe it simply doesn't
work. An example may be the `SourceLocationSequence` optimization.
This will only affect the size of on-disk .pcm files. I don't expect
this impact the speed and memory use of compilations. And seeing my
small experiments above, I feel this trade off is worthy.
# Correctness
The mental model for handling source location offsets is not so complex
and I believe we can solve it by adding module file index to each stored
source location.
For the practical side, since the source location is pretty sensitive,
and the patch can pass all the in-tree tests and a small scale projects,
I feel it should be correct.
# Future Plans
I'll continue to work on no transitive decl change and no transitive
identifier change (if matters) to achieve the goal to stop the
propagation of unnecessary changes. But all of this depends on this
patch. Since, clearly, the source locations are the most sensitive
thing.
---
The release nots and documentation will be added seperately.
This patch tries to remove all the direct use of DeclID except the real
low level reading and writing. All the use of DeclID is converted to
the use of LocalDeclID or GlobalDeclID. This is helpful to increase the
readability and type safety.
This patch tries to remove all the direct use of DeclID except the real
low level reading and writing. All the use of DeclID is converted to
the use of LocalDeclID or GlobalDeclID. This is helpful to increase the
readability and type safety.
Previously, the DeclID is defined in serialization/ASTBitCodes.h under
clang::serialization namespace. However, actually the DeclID is not
purely used in serialization part. The DeclID is already widely used in
AST and all around the clang project via classes like `LazyPtrDecl` or
calling `ExternalASTSource::getExernalDecl()`. All such uses are via the
raw underlying type of `DeclID` as `uint32_t`. This is not pretty good.
This patch moves the DeclID class family to a new header `AST/DeclID.h`
so that the whole project can use the wrapped class `DeclID`,
`GlobalDeclID` and `LocalDeclID` instead of the raw underlying type.
This can improve the readability and the type safety.
Close https://github.com/llvm/llvm-project/issues/87609
We tried to profile the body of the lambda expressions in
https://reviews.llvm.org/D153957. But as the original comments show,
it is indeed dangerous. After we tried to skip calculating the ODR
hash values recently, we have fall into this trap twice.
So in this patch, I choose to not profile the body of the lambda
expression. The signature of the lambda is still profiled.
