fix(emboss): crash in slicing prep after embossing all-space text

[adele-with-a-b]

What breaks

Embossing text that contains only spaces commits a ModelVolume holding an empty mesh. The next background slicing prep then dereferences a null vector data pointer inside Print::apply, taking the application down.

Root cause

GenerateTextJob::process has three early-exit paths that return normally while leaving m_input.m_final_text_mesh empty. finalize() cannot tell that apart from success, so it commits the empty mesh.

Line numbers are against master at 66e405477.

site	condition	function
EmbossJob.cpp:1404	m_chars_mesh_result.empty()	GenerateTextJob::process
EmbossJob.cpp:1935	bbs.empty()	generate_mesh_according_points
EmbossJob.cpp:1940	ms.mesh->empty()	generate_mesh_according_points

Three things make this fatal rather than merely a no-op:

1. generate_mesh_according_points calls mesh.clear() on input_info.m_final_text_mesh at EmbossJob.cpp:1922 before either of its two early exits, so both leave the mesh empty rather than untouched.
2. GenerateTextJob::finalize tests only canceled || eptr (EmbossJob.cpp:1423-1425). It has no emptiness check, so a normal return from process() reads as success and the mesh is committed through create_text_volume() / recreate_model_volume().
3. Print::apply → update_volume_bboxes → transformed_its_bbox2d then evaluates its.vertices[its.indices.front()(0)] (PrintApply.cpp:587). The guarding assert(! its.indices.empty()) one line above is compiled out in release, so front() reads through a null data pointer.

Trigger

All-space text. create_all_char_mesh clears its out-parameter at EmbossJob.cpp:1284 and only afterwards tests for all-space input, at EmbossJob.cpp:1291-1293:

wxRegEx re("^ +$");
bool    is_all_space = re.Matches(input_text);
if (is_all_space) { return; }

Because the clear precedes the early return, m_chars_mesh_result comes back empty rather than unmodified. m_final_text_mesh is then still default-constructed — InputInfo is a stack local (GLGizmoText.cpp:3336) and generate_mesh_according_points is its only writer — so the volume that reaches slicing prep has an empty indexed_triangle_set.

The fix

The three sites now throw JobException instead of returning. That is what every other failure path in these two functions already does — EmbossJob.cpp:1408, 1411, 1415, 1932, and 1964. The last of those is especially relevant: it throws on exactly this condition (mesh.its.empty()) at the end of generate_mesh_according_points. The three sites being changed are the outliers, not a new convention.

No signature changes, no new translatable strings. Both messages already exist in this file for these conditions — "Font doesn't have any shape for given text." at EmbossJob.cpp:1415 and 1932, and "There is no valid surface for text projection." at EmbossJob.cpp:1004 and 1024.

This is the same shape and scope as ef96c2001 ("FIX:A special font does not generate x characters, causing software crashes"), which fixed the neighbouring instance of this bug class in this file.

Why the sibling job doesn't have this bug

CreateObjectTextJob::process has the identical bare return at EmbossJob.cpp:1978, but it is safe: its finalize() guards m_input.m_position_points.empty() before building any geometry (EmbossJob.cpp:1990-1992). That is the contract GenerateTextJob::finalize is missing, and throwing restores it without touching either signature.

Deliberately out of scope

Two adjacent problems are left alone, because bundling either would make this change much harder to reason about:

Hardening update_volume_bboxes against an empty indexed_triangle_set. That would defend against any source of an empty volume, not just this one, and it is worth doing. But it is shared slicing-core code, and the obvious implementation — dropping empty volumes from the vector it walks — is unsafe: update_volume_bboxes is the sole writer of PrintObjectRegions::cached_volume_ids (PrintApply.cpp:955-959), and the consumer at PrintApply.cpp:687-690 indexes that vector in a loop whose condition has no bounds check. Removing volumes from the walk desynchronises the two and turns a deterministic fault into an out-of-bounds read. Doing it properly needs its own change.

Error reporting for these jobs. This stops the crash but does not tell the user anything. GenerateTextJob::finalize and CreateObjectTextJob::finalize are the only two of seven finalize() methods in this file that don't route through _finalize(); the other five (EmbossJob.cpp:350, 373, 456, 535, 552) call if (!_finalize(canceled, eptr, *m_input.base)) return;, which hands the exception to exception_process() and shows the user a message. These two test canceled || eptr and discard eptr, which is also why the five JobExceptions already present in GenerateTextJob::process are silent today. Rewiring them changes user-visible error reporting and belongs in its own change.