Support 'archive: member' file-pattern in input section description

This commit adds support for 'archive: member' pattern in input section
description. Generally, there should be no space between the 'archive:'
and 'member'. However, it is a common mistake to put a space here. This
commit makes this common use-case work as expected and reports a
deprecated warning for the same.

Closes #67

Signed-off-by: Parth Arora <[email protected]>

Author: parth-07

docs/userguide/documentation/linker_faq.rst

@@ -2302,6 +2302,59 @@ into both the :code:`PT_TLS` and a :code:`PT_LOAD` section. For example:
      ...
    }
 
+Why am I getting the warning 'Space between archive:member file pattern is deprecated'
+-----------------------------------------------------------------------------------------
+
+Linker script allows to specify archive members in input section descriptions
+using the `<archive-pattern>:<member-pattern>` syntax. For example in the below
+linker script `SECTIONS` command snippet, `foo_text` output section contains the
+`.text*` sections from the `foo.o` member of `libfoobar.a` archive.
+
+.. code-block:: bash
+
+   SECTIONS {
+     foo_text : { libfoobar.a:foo.o(.text*) }
+   }
+
+
+Please note that there should be no space between the `<archive-pattern>:`
+and the `<member-pattern>`. Previously, the linker accepted the space
+between the two patterns for convenience. This behavior is now
+deprecated and will be removed in the future.
+
+The below linker script snippet demonstrates some of the cases where
+the warning will be emitted and why.
+
+.. code-block:: bash
+
+   cat > 1.c << \!
+   int data = 10;
+   !
+
+   cat > script.t << \!
+   SECTIONS {
+     .data : {
+        /* The warning will be displayed because there is a space between
+           '*lib1.a:' and '1.o' */
+       *lib1.a: 1.o(.*data*)
+        /* No warning. When member-pattern is missing, all members of the
+           matched archives are matched. */
+       /tmp/lib1.a*: (.*data*)
+       *lib1.a: (.*data*)
+       /* The warning will be displayed because there is a space between
+          '*lib1.a:' and '*' */
+       *lib1.a: *(.*data*)
+       /* No warning. '*lib1.a:' and '*(.data*)' are considered as separate
+          input section descriptions. */
+       *lib1.a:
+       *(.data*)
+     }
+   }
+   !
+
+  clang -c   -c 1.c -g
+  llvm-ar cr lib1.a 1.o
+  ld.eld --whole-archive lib1.a -T script.t -Map x -Wlinker-script
 
 Linker Script PHDRS
 ====================

include/eld/Diagnostics/DiagLDScript.inc

@@ -123,3 +123,4 @@ DIAG(warn_non_power_of_2_value_to_align_builtin, DiagnosticEngine::Warning,
 DIAG(error_non_power_of_2_value_to_align_output_section, DiagnosticEngine::Error,
      "%0: non-power-of-2 value 0x%1 passed to ALIGN in '%2' output section description, value must "
      "be 0 or a power of 2")
+DIAG(warn_linker_script, DiagnosticEngine::Warning, "%0")

include/eld/ScriptParser/ScriptLexer.h

@@ -44,6 +44,8 @@ class ScriptLexer {
   void setNote(const llvm::Twine &msg,
                std::optional<llvm::StringRef> columnTok = std::nullopt) const;
 
+  void setWarn(const llvm::Twine &Msg);
+
   void lex();
 
   // Skip spaces
@@ -100,6 +102,12 @@ class ScriptLexer {
   /// is no previous token, this function is no-op.
   void prev();
 
+  size_t computeLineNumber(llvm::StringRef tok);
+
+  size_t getCurrentLineNumber() {
+    return PrevTokLine;
+  }
+
 protected:
   // Get current memory buffer
   llvm::MemoryBufferRef getCurrentMB() const;

lib/ScriptParser/ScriptLexer.cpp

@@ -117,6 +117,14 @@ void ScriptLexer::setNote(const Twine &msg,
   ThisConfig.raise(Diag::note_linker_script) << s;
 }
 
+void ScriptLexer::setWarn(const Twine &Msg) {
+  std::string S = (getCurrentLocation() + ": " + Msg).str();
+  if (PrevTok.size())
+    S += "\n>>> " + getLine().str() + "\n>>> " +
+         std::string(getColumnNumber(), ' ') + "^";
+  ThisConfig.raise(Diag::warn_linker_script) << S;
+}
+
 void ScriptLexer::lex() {
   for (;;) {
     StringRef &S = CurBuf.S;
@@ -145,10 +153,9 @@ void ScriptLexer::lex() {
     if (S.starts_with("\"")) {
       size_t E = S.find("\"", 1);
       if (E == StringRef::npos) {
-        size_t Lineno =
-            StringRef(CurBuf.Begin, S.data() - CurBuf.Begin).count('\n');
+        size_t Lineno = computeLineNumber(S);
         ThisConfig.raise(Diag::error_linker_script)
-            << llvm::Twine(CurBuf.Filename + ":" + Twine(Lineno + 1) +
+            << llvm::Twine(CurBuf.Filename + ":" + Twine(Lineno) +
                            ": unclosed quote")
                    .str();
         return;
@@ -330,6 +337,7 @@ StringRef ScriptLexer::unquote(StringRef S) {
 
 void ScriptLexer::prev() {
   if (!PrevTok.empty()) {
+    // FIXME: CurBuf.LineNumber needs to be updated!
     CurBuf.S = PrevTok.data();
     CurTok = {};
   }
@@ -357,3 +365,9 @@ size_t ScriptLexer::computeColumnWidth(llvm::StringRef s,
   });
   return e.data() - s.data() - nonASCIIColumnOffset;
 }
+
+size_t ScriptLexer::computeLineNumber(llvm::StringRef tok) {
+  size_t LineNumber =
+      llvm::StringRef(CurBuf.Begin, tok.data() - CurBuf.Begin).count('\n');
+  return LineNumber + 1;
+}

lib/ScriptParser/ScriptParser.cpp

@@ -684,8 +684,17 @@ InputSectDesc::Spec ScriptParser::readInputSectionDescSpec(StringRef Tok) {
   else {
     std::pair<llvm::StringRef, llvm::StringRef> Split = Tok.split(':');
     FilePat = ThisScriptFile.createWildCardPattern(Split.first);
-    if (!Split.second.empty())
+    if (!Split.second.empty()) {
       ArchiveMem = ThisScriptFile.createWildCardPattern(Split.second);
+    }
+    llvm::StringRef peekTok = peek();
+    if (!atEOF() && peekTok != "(" &&
+        computeLineNumber(peekTok) == getCurrentLineNumber()) {
+      next();
+      if (ThisConfig.showLinkerScriptWarnings())
+        setWarn("Space between archive:member file pattern is deprecated");
+      ArchiveMem = ThisScriptFile.createWildCardPattern(peekTok);
+    }
     IsArchive = true;
   }
   StringList *WildcardSections = nullptr;

test/Common/standalone/linkerscript/ArchiveFilePatternWithSpace/ArchiveFilePatternWithSpace.test

@@ -0,0 +1,39 @@
+#---ArchiveFilePatternWithSpace.test--------------------- Executable ------------------#
+#BEGIN_COMMENT
+# This tests that checks that 'archive: file' pattern works for now and a
+# deprecated diagnostic is emitted for this usage.
+#END_COMMENT
+RUN: %clang %clangopts -o %t1.1.o %p/Inputs/1.c -c -ffunction-sections
+RUN: %clang %clangopts -o %t1.2.o %p/Inputs/2.c -c -ffunction-sections
+RUN: %ar cr %t1.lib12.a %t1.1.o %t1.2.o
+RUN: %link %linkopts -o %t1.a1.out --whole-archive %t1.lib12.a \
+RUN:   -T %p/Inputs/script.1.t -Map %t1.a1.map.txt -Wno-linker-script 2>&1 | %filecheck %s --allow-empty --check-prefix NO_WARN
+RUN: %link %linkopts -o %t1.a1.out --whole-archive %t1.lib12.a \
+RUN:   -T %p/Inputs/script.1.t -Map %t1.a1.map.txt -Wlinker-script 2>&1 | %filecheck %s --check-prefix WARN
+RUN: %filecheck %s --check-prefix MAP < %t1.a1.map.txt
+RUN: %link %linkopts -o %t1.a2.out --whole-archive %t1.lib12.a \
+RUN:   -T %p/Inputs/script.2.t -Map %t1.a2.map.txt
+RUN: %filecheck %s --check-prefix MEMBER_PATTERN_NEXT_LINE < %t1.a2.map.txt
+
+NO_WARN-NOT: Space between
+
+WARN: Warning: {{.*}}script.1.t:2: Space between archive:member file pattern is deprecated
+WARN: >>>   FOO: { *lib12.a: *1.o(*foo*) }
+WARN: >>>                    ^
+WARN: Warning: {{.*}}script.1.t:3: Space between archive:member file pattern is deprecated
+WARN: >>>   BAR: { *lib12.a: *(*bar*) }
+WARN: >>>
+
+MAP: FOO
+MAP: *lib12.a:*1.o(*foo*)
+MAP: .text.foo
+
+MAP: BAR
+MAP: *lib12.a:*(*bar*)
+MAP: .text.bar
+
+MEMBER_PATTERN_NEXT_LINE: WRONG_DATA
+MEMBER_PATTERN_NEXT_LINE: *lib12.a:(*)
+MEMBER_PATTERN_NEXT_LINE: .text.foo
+MEMBER_PATTERN_NEXT_LINE: .text.bar
+MEMBER_PATTERN_NEXT_LINE: *(*data*)

test/Common/standalone/linkerscript/ArchiveFilePatternWithSpace/Inputs/1.c

@@ -0,0 +1 @@
+int foo() { return 1; }

test/Common/standalone/linkerscript/ArchiveFilePatternWithSpace/Inputs/2.c

@@ -0,0 +1 @@
+int bar() { return 3; }

test/Common/standalone/linkerscript/ArchiveFilePatternWithSpace/Inputs/script.1.t

@@ -0,0 +1,4 @@
+SECTIONS {
+  FOO: { *lib12.a: *1.o(*foo*) }
+  BAR: { *lib12.a: *(*bar*) }
+}

test/Common/standalone/linkerscript/ArchiveFilePatternWithSpace/Inputs/script.2.t

@@ -0,0 +1,6 @@
+SECTIONS {
+  WRONG_DATA: {
+    *lib12.a:
+      *(*data*)
+  }
+}