Move declaration check out and do it in a separate step

[lvjr]

In fixing #22, \debug_on:n { check-declarations } was added to table-001.tex. But it brings a problem that the test will fail whenever I add a new variable. This is not that good. Maybe we need to move declaration check out and do it in a separate step.

[lvjr]

Or to find a way to remove these info lines from table-001.tlg file.

\l__tblr_a_int=\count...
\l__tblr_c_int=\count...
\l__tblr_r_int=\count...
\l__tblr_d_dim=\dimen...
\l__tblr_h_dim=\dimen...
\l__tblr_o_dim=\dimen...
\l__tblr_p_dim=\dimen...
\l__tblr_q_dim=\dimen...
\l__tblr_r_dim=\dimen...
\l__tblr_s_dim=\dimen...
\l__tblr_t_dim=\dimen...
\l__tblr_v_dim=\dimen...
\l__tblr_w_dim=\dimen...
\l__tblr_a_box=\box...
\l__tblr_b_box=\box...
\l__tblr_c_box=\box...
\l__tblr_d_box=\box...
\g__tblr_table_count_int=\count...
\c@rownum=\count...
\c@colnum=\count...
\c@rowcount=\count...
\c@colcount=\count...
\abovesep=\dimen...
\belowsep=\dimen...
\leftsep=\dimen...
\rightsep=\dimen...
\g_tblr_level_int=\count...
\g__tblr_data_row_key_count_int=\count...
\g__tblr_data_column_key_count_int=\count...
\g__tblr_data_cell_key_count_int=\count...
\g__tblr_array_int=\count...
\l__tblr_key_count_int=\count...
\l__tblr_key_quotient_int=\count...
\l__tblr_key_quotient_two_int=\count...
\l__tblr_key_remainder_int=\count...
\g__tblr_data_str_value_count_int=\count...
\rulewidth=\dimen...
\l__tblr_strut_dp_dim=\dimen...
\l__tblr_strut_ht_dim=\dimen...
\g__tblr_cell_wd_dim=\dimen...
\g__tblr_cell_ht_dim=\dimen...
\g__tblr_cell_head_dim=\dimen...
\g__tblr_cell_foot_dim=\dimen...
\l__tblr_column_target_dim=\dimen...
\l__tblr_caption_box=\box...
\l__tblr_caption_left_box=\box...
\l__tblr_row_head_box=\box...
\l__tblr_row_foot_box=\box...
\l__tblr_row_head_foot_dim=\dimen...
\tablewidth=\dimen...
\l__tblr_table_firsthead_box=\box...
\l__tblr_table_middlehead_box=\box...
\l__tblr_table_lasthead_box=\box...
\l__tblr_table_firstfoot_box=\box...
\l__tblr_table_middlefoot_box=\box...
\l__tblr_table_lastfoot_box=\box...
\l__tblr_remain_height_dim=\dimen...
\l__tblr_long_from_int=\count...
\l__tblr_long_to_int=\count...
\l__tblr_curr_i_int=\count...
\l__tblr_prev_i_int=\count...
\l__tblr_table_page_int=\count...
\l__tblr_table_head_box=\box...
\l__tblr_table_foot_box=\box...
\l__tblr_table_head_foot_dim=\dimen...
\l__tblr_table_head_body_foot_dim=\dimen...
\l__tblr_table_box=\box...
\l__tblr_table_hlines_box=\box...
\l__tblr_hline_box=\box...
\l__tblr_row_box=\box...
\l__tblr_col_o_wd_dim=\dimen...
\l__tblr_col_b_wd_dim=\dimen...
\l__tblr_hline_leftskip_dim=\dimen...
\l__tblr_hline_rightskip_dim=\dimen...
\l__tblr_row_ht_dim=\dimen...
\l__tblr_row_dp_dim=\dimen...
\l__tblr_row_abovesep_dim=\dimen...
\l__tblr_row_belowsep_dim=\dimen...
\l__tblr_row_vlines_box=\box...
\l__tblr_vline_box=\box...
\l__tblr_cell_box=\box...
\l__tblr_row_upper_dim=\dimen...
\l__tblr_row_lower_dim=\dimen...
\l__tblr_row_vpace_dim=\dimen...
\l__tblr_vline_aboveskip_dim=\dimen...
\l__tblr_vline_belowskip_dim=\dimen...
\l__tblr_cell_wd_dim=\dimen...
\l__tblr_cell_ht_dim=\dimen...
\l__tblr_diag_box=\box...

[muzimuzhi]

But it brings a problem that the test will fail whenever I add a new variable.

Only registers are accounted. Tl-based variables don't add log lines.

So the question is, would the registers in-use be relatively stable?

[lvjr]

You can't expect private registers to be stable. And even shifting the definition of a public register would make the test fail.

[lvjr]

I decide to move \debug_on:n { check-declarations } to manual.sty. In this way we can still catch the errors before new releases.

Also we may add lualatex manual-0.tex to texlive.yml. It seems pgf builds its manual in regression tests too.

[lvjr]

By the way, I found two more experimental public variables \rulewidth and \tablewidth in the above log file and have added them to issue #527.

[lvjr]

@muzimuzhi I didn't add lualatex manual-0.tex to texlive.yml. Not sure whether or not you like it, so I leave it to you.

[muzimuzhi]

It seems pgf builds its manual in regression tests too.

pgf doesn't have comprehensive test suite(s), so it compiles its example-rich manual as a (temporary) way to do regression tests. See pgf-tikz/pgf#1116 for its future plan.

Not sure whether or not you like it, so I leave it to you.

No strong opinion here. Feel free.

[muzimuzhi]

Note that \debug_on:n (or \sys_load_debug:) loads l3debug.def which wraps most l3kernel functions. This wrapping cannot be reverted by \debug_off:n so the patched l3kernel functions are permanently made a bit slower.

This may slow down compilation of the package manual

Compiling manual as part of the regression test is similar to setting recordstatus = true for l3build tests (also mentioned in #525 (comment)).

With recordstatus = true, lines like

***************
Compilation 1 of test file completed with exit status 0

will be appended to .tlg files.

One drawback is that l3build doesn't support per-testfile configs, so we cannot turn on recordstatus for a single test file without creating a new test set.

[lvjr]

Note that \debug_on:n (or \sys_load_debug:) loads l3debug.def which wraps most l3kernel functions. This wrapping cannot be reverted by \debug_off:n so the patched l3kernel functions are permanently made a bit slower.

Then I think it is a l3debug bug.

[lvjr]

As a workaround of the l3debug bug, I decide to only check declarations in single-chapter compilation.

[lvjr]

A future plan is to make codehigh run faster with LuaTeX (it is possible if we use pure Lua).