Tables not consistently full width

[alerque]

I'm using a scripted process to generate reports on financial data that includes a lot of tables. Mostly this is done be generating CSV files and then a Markdown file that includes them. There are a couple different table layouts, but each layout is used several times. The code I'm using generates consistent Markdown, but the output is entirely irregular and I can't figure out why.

All tables have meta data that specifies the table width and column widths. As the code runs a for look to generate tables from a series of CSV data sets I would expect all the tables to render the same. They don't.

For example check out these results. First lets inspect the generated Markdown source file. I'm greping it for width related YAML keys found only in the meta data of tables.

$ cat status.md | grep width: | head -n18
table-width: 1.0
width: [0.2, 0.8]
table-width: 1.0
width: [0.2, 0.2, 0.2, 0.4]
table-width: 1.0
width: [0.2, 0.8]
table-width: 1.0
table-width: 1.0
width: [0.06, 0.64, 0.1, 0.2]
table-width: 1.0

Note that of the first 10 tables in the source, all of them should be 100% width. Table 4 does not specify column widths and leaves it to auto-align. Tables 1 & 3 should both have the same column widths, as should 5 & 6 and 7 & 8 & 9. This is not what happens. When rendering to PDF via LaTeX the tables look like this on the page:

Note that neither of the tables are drawn full width. The same happens for the other sets. Tables 6 and 8 are both actually 100%, but 5 and 7 which should have similar layouts are both somewhat less.

Outputting to HTML tells a similar story. Looking at the table definitions of the first 10 tables output:

$ cat status.html | grep '<table' | head -n10
<table style="width:76%;">
<table>
<table style="width:50%;">
<table>
<table style="width:94%;">
<table>
<table style="width:100%;">
<table style="width:100%;">
<table>
<table style="width:100%;">

A few of these tables are left to their own means for width. A few of them have 100% definitions. The others have bizarre specific numbers that come out of nowhere. Note the 76% and 50% widths for tables 1 and 3 respectively correspond to what the LaTeX output did as well.

Here is the full source code at the start of table 1 and 3 for comparison and to show there isn't something else funky going on in the source. Only the CSV format data is being obscured here:

How do I end these shenanigans and just get full width tables all the time?

[alerque]

As an experiment instead of going directly to PDF I had Pandoc write a standalone LaTeX format output. The code starting table 1 looks like this:

\begin{longtable}[]{@{}rl@{}}
\toprule
\begin{minipage}[b]{0.20\columnwidth}\raggedleft
balance\strut
\end{minipage} & \begin{minipage}[b]{0.52\columnwidth}\raggedright
account\strut
\end{minipage}\tabularnewline
\midrule
\endhead

The first column being 0.20\columnwidth sounds correct to me, but where it is coming up with the second column (specificed as 0.8 in the Pantable meta data) should be 0.52\columnwidth I have no idea. Given that the intro for table 3 is completely different, it would seem the 0.2 matching up is just a fluke.

\begin{longtable}[]{@{}rl@{}}
\toprule
\begin{minipage}[b]{0.18\columnwidth}\raggedleft
balance\strut
\end{minipage} & \begin{minipage}[b]{0.29\columnwidth}\raggedright
account\strut
\end{minipage}\tabularnewline
\midrule
\endhead

[alerque]

This seems to be a side effect of my workaround for #42. Because I can't both include LaTeX code in the document header and use Pantable I was invoking Pandoc twice, once to render the CSV tables to Markdown and a second time to render the final outputs. Since Markdown doesn't have any comprehension of table widths that information was being lost.