tree 59a23c0d679595a022f8c42247bcab04beb1cb36
parent 77781b0be4c9540cdf3cf1abfa728ad74d0f94ef
author Jiang Ning <jiangninghx@foxmail.com> 1778683207 +0800
committer GitHub <noreply@github.com> 1778683207 +0100
gpgsig -----BEGIN PGP SIGNATURE-----
 
 wsFcBAABCAAQBQJqBI1HCRC1aQ7uu5UhlAAASpUQAIq3NMdri9dOIT5BCwsFeJiC
 +J/2vyRt7SxemxH1XJFueFzMHCXm6SrOJ7JYE1RqW5LWEUz4+MCVkrPrEceolL+B
 D0XdIJUs/xD7JJCB7g7U+GrWGFk9OcOjTydv27Z+wlRKrPZFmH3snzsFCC0bnAm5
 FdDXaTtB84baKd+LY2EqTffTpRsJS7nxgywfZHRz2qUyLH4eezm981ZJ/WRyJvxo
 3MfzEuB2iBTcs8xPnWX5E45aOjtmTg0qBtuF7Sd/FsUlnG0QxEbF1Q9PYOr4JMqq
 bRcY/Djlz2YXwpuzsOUpso+gYPnZHHqu9+bqcyrY0kDuATpkwrC1VnvKKeiN8CRK
 0OyOEc9VAnJiltbwY1N8LGGdt4hO9UmYEUgi9aFUO6W/lmT52adu75aee7vOaS9k
 MM03iJiVkalOCCCSJBvxjkhPINSLK9Iz2Mb1z/Ejjiuo9sVC+jYKTEZcF4ZzAS1T
 sRkJksmUtVtViF8BjfK3n/dzRkUqT7+2PdPTzxdWDSYKRc5+TCHF6+qZudjpr76S
 a3qrdiuRJntU18vPs23A2Y6UbmDPBuD1c3F1oi+0ebwoWvDoEBaxP4y+lFkqgPAQ
 FXeOAF+RNzkVeaVwn+gz7qMm3Qe6lUkgsObeeyVN+/4p5e4HW+aEa4Z59x3wLPLx
 MTf2oKoYMZu959Z3cGKn
 =Ga7X
 -----END PGP SIGNATURE-----
 

[DebugInfo] Enforce DISubprogram type as required in LLParser (#196299)

Add a parse-time check so that LLParser treats the `type:` field of
`DISubprogram` as required and non-null. Attempting to assemble IR with
  a missing or null `type:` now produces a clear parse-time error
("missing required field 'type'" or "'type' cannot be null") rather
than relying solely on the verifier warning added in #194556.
Update existing hand-written LLVM IR tests to supply a minimal valid
`DISubroutineType` where they are not intentionally testing malformed
debug info.
**Scope.** This patch covers the LLParser path only. Two entry points
are intentionally deferred:
- *MetadataLoader*: pre-LLVM-5 bitcode legitimately stores null
    `DISubprogram.type`; auto-upgrading that field requires more careful
backward-compatibility handling (see
`llvm/test/Bitcode/dityperefs-3.8.ll`).
- *BitcodeWriter*: IR written to bitcode has already passed the
verifier, which guarantees a non-null type; no additional check is
needed in the writer.
  Follow-up to #194556.