Failling to build LTS versions with ICU 59

[ArchangeGabriel]

When trying to build either Argon or Boron with ICU 59.x, it fails with the following error:

../deps/v8/src/runtime/runtime-i18n.cc: In function ‘v8::internal::Object* v8::internal::__RT_impl_Runtime_StringNormalize(v8::internal::Arguments, v8::internal::Isolate*)’:
../deps/v8/src/runtime/runtime-i18n.cc:588:8: error: ‘icu::Normalizer’ has not been declared
   icu::Normalizer::normalize(u_value, normalizationForms[form_id], 0, result,
        ^~~~~~~~~~

I’m not sure how to fix this, and if other changes are required, but this is an issue for ArchLinux packages.

[addaleax]

/cc @srl295

[refack]

Ref: bump ICU to 59.1, I'm not sure it was backported to v6 (Boron), and it will probably won't be ported to v4 that has moved from LTS to maintenance.

Ref 2: Previous PR depends on #11753

[ArchangeGabriel]

Hum, maybe I wasn’t clear but I’m building with system ICU, not nodejs embedded one, so #12486 seems irrelevant. I’ve tried to use #11753 as patch, but it didn’t worked. The patching has to be different in v6 and v4 than in v7, and that fixing one (v6) would probably fix both.

[bnoordhuis]

That specific build error is probably easy to fix by adding a #include "unicode/normlzr.h" at the top of runtime-i18n.cc.

In general though, V8 only supports one ICU version at a time, and with the v4.x and v6.x release lines that is ICU 58.

[ArchangeGabriel]

@bnoordhuis Thanks, this fixed it. Build fine, only 3 tests not passing:

=== release test-intl ===                                                      
Path: parallel/test-intl
/build/nodejs-lts-argon/src/node-v4.8.3/test/parallel/test-intl.js:44
      new Intl.DateTimeFormat(['en'],
      ^

RangeError: Unsupported time zone specified UTC
    at Object.<anonymous> (/build/nodejs-lts-argon/src/node-v4.8.3/test/parallel/test-intl.js:44:7)
    at Module._compile (module.js:409:26)
    at Object.Module._extensions..js (module.js:416:10)
    at Module.load (module.js:343:32)
    at Function.Module._load (module.js:300:12)
    at Function.Module.runMain (module.js:441:10)
    at startup (node.js:140:18)
    at node.js:1043:3
Command: out/Release/node /build/nodejs-lts-argon/src/node-v4.8.3/test/parallel/test-intl.js
=== release test-tls-ecdh-disable ===                                          
Path: parallel/test-tls-ecdh-disable
assert.js:81
  throw new assert.AssertionError({
  ^
AssertionError: false == true
    at /build/nodejs-lts-argon/src/node-v4.8.3/test/parallel/test-tls-ecdh-disable.js:34:5
    at /build/nodejs-lts-argon/src/node-v4.8.3/test/common.js:407:15
    at ChildProcess.exithandler (child_process.js:207:5)
    at emitTwo (events.js:87:13)
    at ChildProcess.emit (events.js:172:7)
    at maybeClose (internal/child_process.js:862:16)
    at Socket.<anonymous> (internal/child_process.js:338:11)
    at emitOne (events.js:77:13)
    at Socket.emit (events.js:169:7)
    at Pipe._onclose (net.js:490:12)
Command: out/Release/node /build/nodejs-lts-argon/src/node-v4.8.3/test/parallel/test-tls-ecdh-disable.js
=== release test-tls-set-ciphers ===                                           
Path: parallel/test-tls-set-ciphers
/build/nodejs-lts-argon/src/node-v4.8.3/test/parallel/test-tls-set-ciphers.js:45
    if (err) throw err;
             ^

Error: Command failed: "openssl" s_client -cipher DES-CBC3-SHA -connect 127.0.0.1:44101
Error with command: "-cipher DES-CBC3-SHA"
140587604950912:error:1410D0B9:SSL routines:SSL_CTX_set_cipher_list:no cipher match:ssl/ssl_lib.c:2018:

    at ChildProcess.exithandler (child_process.js:200:12)
    at emitTwo (events.js:87:13)
    at ChildProcess.emit (events.js:172:7)
    at maybeClose (internal/child_process.js:862:16)
    at Socket.<anonymous> (internal/child_process.js:338:11)
    at emitOne (events.js:77:13)
    at Socket.emit (events.js:169:7)
    at Pipe._onclose (net.js:490:12)
Command: out/Release/node /build/nodejs-lts-argon/src/node-v4.8.3/test/parallel/test-tls-set-ciphers.js

The DES-CBC3-SHA one is because although nodejs is built against OpenSSL 1.0, the openssl command on my system is OpenSSL 1.1, so this cipher is gone, hence the error. Not sure about the other two tests, but I would expect the test-tls-ecdh-disable to be something similar. No idea about the first one.

[ArchangeGabriel]

Currently testing on my server, this single line change seems to be enough. Should I expect something to break, and if so how to detect any breakage?

[bnoordhuis]

The parallel/test-intl failure is probably related. As to breakage: it's an untested combination; bugs are possible.

I think we'd be open to floating a patch to fix the build error, open a pull request if you want to purse that. I'll go ahead and close this out.

[ArchangeGabriel]

Should I open PR against -staging branches or just v4.x/v6.x ones?

[bnoordhuis]

The -staging ones. The -staging branches contain what is to be released, the non-staging ones what has been released.

[ArchangeGabriel]

OK, thanks.

[ArchangeGabriel]

I’ve opened #13040. I’ll open one for v6.x once I’ll have fixed this one enough for it to be accepted.

[Qantas94Heavy]

@ArchangeGabriel I've just tried building 4.x with ICU 59 myself and also saw this failure for parallel/test-intl. It seems to be a bug in V8 (fixed in later versions) -- wonder if it's worth fixing: https://bugs.chromium.org/p/chromium/issues/detail?id=364374

[srl295]

late once again.

Yes, as @MylesBorins commit says, the #include above is a latent v8 bug, exactly like #11753 . (I think the Normalizer is the deprecated API anyway.)

[srl295]

In general though, V8 only supports one ICU version at a time, and with the v4.x and v6.x release lines that is ICU 58

Previous ICUs should be supported as well, at this point.

[srl295]

@Qantas94Heavy but, your assessment of the TZ part seems correct.