<table border="1" cellspacing="0" cellpadding="8">
    <tr>
        <th>Issue</th>
        <td>
            <a href=https://github.com/llvm/llvm-project/issues/104734>104734</a>
        </td>
    </tr>

    <tr>
        <th>Summary</th>
        <td>
            FR: Have -fno-exceptions cause noexcept operator/std::is_nothrow_...() to treat functions as if they were non-throwing
        </td>
    </tr>

    <tr>
      <th>Labels</th>
      <td>
            new issue
      </td>
    </tr>

    <tr>
      <th>Assignees</th>
      <td>
      </td>
    </tr>

    <tr>
      <th>Reporter</th>
      <td>
          pkasting
      </td>
    </tr>
</table>

<pre>
    Chromium compiles (everything, including the STL) with `-fno-exceptions`. However, tests like the `noexcept` operator or `std::is_nothrow_move_constructible<>()` still treat functions as potentially-throwing unless they are explicitly marked `noexcept`. This means authors must still mark various places (e.g. move constructors of objects potentially stored in vectors) as `noexcept` to avoid potential pessimizations.

I believe it's also potentially pessimizing to blindly mark every function as `noexcept`, especially if functions we author call into third-party code not so marked; IIRC, the compiler has to emit code to handle potential exceptions from such code? When I discussed this a few years ago with some LLVM person (I forget who), I think this was a reason that we wouldn't want `-fno-exceptions` to actually compile code equivalent to if the coder had written `noexcept` on every declaration. (Sorry, memory is fuzzy; I just remember there was a good reason to not do "the simple thing" here.)

One possible way to have our cake and eat it too would be to have `-fno-exceptions` cause the `noexcept` operator, along with builtins like `__is_nothrow_constructible()` (which I think are what libc++ implements the STL in terms of), always claim that functions so queried were, indeed, non-throwing. Presumably this would lead to crashes or worse if code compiled this way was linked against a library not compiled this way; we'd be OK with that.

Hopefully, this would result in things like `constexpr if` fast-paths for non-throwing code always being taken, without the downsides of actually treating all contexts as `noexcept`, and without authors having to remember to so tag code.
</pre>
<img width="1px" height="1px" alt="" src="http://email.email.llvm.org/o/eJx8VV9v2zgM_zTKC1HDtZO0echD212w4nbYYRvuHgvZomOtsuSJdFzv0x8ou2m7FffSBrBIkb9_0kT26BH3anOrNh9WeuA2xH3_qImtP66qYKb9XRtDZ4cO6tD11iGBKq7xhHHi1vqjKu7A-toNxvojcIvw9dsnVexgtNyC2uYXjQ8X-FRjzzZ4Uts8g49hlA5Sy0hM4OwjpmK1zX2YT6ttDqHHqDlECFE-ERtV3qjyxtKDD9zGMD504YQPdfDEcajZVg5VeafKP1RxrYqdNCG2zgFH1AzN4Os0B2iCPjB6ttq56SI1kxUG75BIhplARwR86p2tLbsJOh0f0fwyYwbfWkvQoZamCUGCbiBe7pUiOOlow0DQO10vCGbHDGR2OM8uhaGBUH3Hmt9MB8QhogHr4YTpoCCs6Ve4OIA-BWteaqFHItvZnzptnan8g8pv5r_3UKGzeEKwrIorAu0ovLn2uThRG6By1psFB0gSOOP5-zBCLlKP9dzKNq-wH3FBCmrtHFjPAbi10Vz0OvIEdTAIPjBQWEBX5S3c33-5S5Jp8VmMEVpNMhp2lucyDtBqbxy-AuFFftDE0AENdZtOq_IA_7bo4R6MpXogQiOTEGhocIQJdSTQxzDLmUKH8OnTP39Bj5GCFyLvoQnxiAxjG0RwxR3cSwv_ODcatTSLqOU8t5pl-zEMznhVXDGM2vP7Pkl01jwk_JaF5x3xx2BP2qFnOWObBRKT8DAwRsuM_jcz-YU0g7XTMSkikxW-hhgnGbzDLsQJLEEz_Pw5JdDhu2g5YoddhVFuirgsdQzBnDcLiTATQBWFjEO26x3CEhIFSF0mAL1S4GcvLBGJa2HU00zeCSEMIo1HBO0NiG-tbBpm3KDC88H3gav1QP-bJ7KsdsEfZ16rwTq2fskhtc0fHl4lzNtwOceKKq7H1tbtmW5Ji1EIdraqVXGriltIIHTomZ6zUUzMGDvx-qIX7UY9EdRO226WyItVKMCPAaNFAyNGnOPWIBr55YM_B1cGf0ekodOVmxblJbAcaiNw1VFTiyRJOoZIKLJJYlqUZZ7lOiV2nfWSdfqorScGLUtFHafE8m8lopQRVXGVyPn85wyrrPImcT6GHpvBuWm28XlGGdxxQkbk8sJDgh6f-gi2EcwbTXzRa25JXPdm_3mZBcoKU2TpR_RylUwTBk4UmDB6sgZT1p7tlV4HqZE8qoNnfGJ6P9REks8Nn_O-1aclI1-MEoQ71vNg2crsS7Mrd3qF-8urYp2Xu8vdetXucV01eoubjcl1tb3Ksdjlu6rZ5s315W6z3q3svsiLdX59ucvL9bZcZ42uN7uy2tTralOaSqt1jp22LnPu1GUhHleWaMD9Zb6-Ktcrpyt0lN74ovA4QvqqikKe_LiXootqOJJa584S00sbtuxwf_iiyhv4KG77xWqLz57xeeWuw3svdZZls3sEm3ee4znHpqTzN9Suhuj2LXNP0rM4qOJwtNwOVVaHThUHGXj5d9HHIM-nKg5pTVLFYcHhtC_-CwAA__8txBzS">