erb-format is deleting text from the copy of a template file

[jbranchaud]

I ran into this in a codebase and was able to put together this reproduction.

$ erb-format --version
ERB::Formatter 0.7.3

Here is the repro template where Case 1 gets mangled by the formatter whereas Case 2 and Case 3 are untouched:

<%# Test data inline
checkout =
  OpenStruct.new(
    status: :overdue,
    loan_period: OpenStruct.new(name: "Standard"),
  )

next_fine =
  OpenStruct.new(assessed_date: Date.new(2024, 3, 15), amount_owed: 12.50) %>

<div class="patron-details">
  <!-- Case 1: Text "fine" after dot character -->
  <div>
    <%= checkout.status %>
    - fine
    <%= next_fine.assessed_date.strftime("%b %d") %>
    @ branch •
    <%= checkout.status == :overdue ? "Current" : "Pending" %> fine as of
    <%= next_fine.assessed_date.strftime("%b %d") %> totaling
    <%= number_to_currency next_fine.amount_owed %>
  </div>
  <!-- Case 2: Text "fine" after wrapped dot character -->
  <div>
    <%= checkout.status %>
    - fine
    <%= next_fine.assessed_date.strftime("%b %d") %>
    @ branch
    <%= "•" %>
    <%= checkout.status == :overdue ? "Current" : "Pending" %>
    fine as of
    <%= next_fine.assessed_date.strftime("%b %d") %>
    totaling
    <%= number_to_currency next_fine.amount_owed %>
  </div>
  <!-- Case 3: Extra minimal -->
  <div>
    Text •
    <%= checkout.status == :overdue ? "Current" : "Pending" %>
    fine as of
  </div>
</div>

Here is what the diff looks like:

What is perhaps odder is that subsequent runs of the formatter against the "formatted" file then remove additional pairs of characters.

[jbranchaud]

Oh my, multi-codepoint emojis really do a number on it. I think this is a case of StringScanner#pos vs StringScanner#charpos.