html/template: unexpected semantic differences compared to text/template

[rsc]

I ported the text/template tests over to html/template and discovered some semantic differences between the two that I did not expect.

In html/template:

• Missing values format as empty strings, while in text/template they format as <no value>. For example:

  • {"map .NO", "{{.MSI.NO}}", "", tVal, true},

  • {"map .NO", "{{.MSI.NO}}", "", tVal, true}, // NOTE: in text/template

  • {"empty nil", "{{.Empty0}}", "", tVal, true},

  • {"empty nil", "{{.Empty0}}", "", tVal, true}, // NOTE: in text/template

    and so on

• Missing values passed explicitly to the html escaper format as <nil> instead of <no value>:

  • {"html untyped nil", {{html .Empty0}}, "<no value>", tVal, true},
  • {"html untyped nil", {{html .Empty0}}, "<nil>", tVal, true}, // NOTE: "<no value>" in text/template

• Given an addressable struct field with a pointer-receiver String method, the String method is not used (in text/template it is):

  • {"V{6666}.String()", "-{{.V0}}-", "-<6666>-", tVal, true},
  • {"V{6666}.String()", "-{{.V0}}-", "-{6666}-", tVal, true}, // NOTE: -<6666>- in text/template

• Given an addressable struct field with a pointer-receiver Error method, the Error method is not used (in text/template it is):

  • {"W{888}.Error()", "-{{.W0}}-", "-[888]-", tVal, true},
  • {"W{888}.Error()", "-{{.W0}}-", "-{888}-", tVal, true}, // NOTE: -[888] in text/template

• TestExecuteOnNewTemplate crashes on all the cases from #11379.

• TestMessageForExecuteEmpty shows a different error message:

  • template: empty: "empty" is an incomplete or empty template
  • template: "empty" is an incomplete or empty template

• TestIssue31810 fails (“check that a parenthesized first argument behaves properly”, for #31810).

• TestTemplateLookUp (for #10910, #10926) fails. Lookup returns non-nil values for undefined templates.

• Calling Parse("") does not overwrite the text of a previous template parse, as it does in text/template. See TestEmptyTemplate, specifically the last case in the table:

  {[]string{"{{.}}", ""}, "twice", ""},

The <no value> change looks like it may possibly be intentional. All the others look like bugs.

/cc @robpike

[gopherbot]

Change https://golang.org/cl/241084 mentions this issue: html/template: add tests from text/template

[cfiderer]

Hi Go experts, is there any news on these behavioral differences? We'd like to have missing values rendered as "<no value>" in HTML templates, too...

[rsc]

@cfiderer My guess is that at this point adding <no value> to HTML templates would break Hugo users quite badly. /cc @bep

In general I think it would make sense to track down why these differences exist before removing them.