As a mechanism

Sarrasine interrupts La Zambinella’s confession that she is a castrato:

“I can give you no hope,” she said. “Cease to speak thus to me, for they would make fool of you. It is impossible for me to shut the door of the theatre to you; but if you love me, or if you are wise, you will come there no more. Listen, monsieur…” she said in a low voice.
“Oh, be still!” said the impassioned artist. “Obstacles make my love more ardent.”
-Balzac, Sarrasine

Barthes comments:

If we have a realistic view of character, if we believe that Sarrasine has a life off the page, we will look for motives for this interruption (enthusiasm, unconscious denial of the truth, etc.). If we have a realistic view of discourse, if we consider the story being told as a mechanism which must function until the end [my italics], we will say that since the law of narrative decrees that it continue, it was necessary that the word castrato not be spoken.
-S/Z, LXXVI

We have all felt this at some point. Characters act improbably because of narrative requirements. But rather than cynically turn our gaze to the writer, and just say that this problem stems from the requirements of composition, the structuralist allows the literary system to absorb the problem, folding the contradiction back into its text—the “common sentence.” Thus Barthes goes on to say say that these two views “support each other:”

A common sentence is produced which unexpectedly contains elements of various languages [my italics]: Sarrasine is impassioned because the discourse must not end; the discourse can continue because Sarrasine, impassioned, talks without listening… From a critical point of view, therefore, it is as wrong to suppress the as it is to take him off the page in order to turn him into a psychological character (endowed with possible motives): the character and the discourse are each other’s accomplices.
-S/Z, LXXVI

But is particularly structuralist is that this text is still an effect of discourse (this, by the way, is what distinguishes Derrida's text from Barthes). In other words, critical analysis is always on the side of discourse:

Such is discourse: if it creates characters, it is not to make them play among themselves before us but to play with them, to obtain from them a complicity which assures the uninterrupted exchange of the codes: the characters are types of discourse and, conversely, the discourse is a character like the others.
-S/Z, LXXVI

Discourse wins, despite the reciprocity. Otherwise characters would play among themselves, in an imaginary literary world. The mechanism keeps working, and as it absorbs contradictions it also absorbs the “realistic view of character.” And rightly so, though this announces a limit to structuralism of sorts: we can’t look at character as some sort of motivated, coherent subject. Why? Because this completely fails to grasp the literary system as literary—that is, fictional. Furthermore, it does not allow us to distinguish basic functions: for example, there is a distinct difference between the improbable act of a character due to the needs of the story, and an improbable act of a character that actively delays the unfolding of a story. The first is what is under discussion here, as we consider the literary system “which must function until the end.” The second forms a subset of the first: it is precisely an effect of this functioning, which Barthes rightly calls a unit of the hermeneutic code. (It is notable that “end” in Barthes statement then does not mean merely a temporal end: it is—in accordance with structuralist notions of finitude—a spatial limit. Schlovsky, for example, cannot distinguish between the two.)

Thus, character is a tying together of various strands of the text, even if viewing them as personalities is somewhat legitimate. Because the looking at the text qua text can absorb this psychologistic view, however, as far as structuralism is concerned, there is no question who is ultimately the authority: the psychologistic view has no comparable ability to explain the textual phenomena except by evading the matter and cynically going to the author or denying fictionality. Barthes can sum up the structuralist view of character quite simply as follows:

When identical semes traverse the same proper name several times and appear to settle upon it, a character is created. Thus, the character is a product of combination: the combination is relatively stable (denoted by the recurrence of the semes) and more or less complex (involving more or less congruent, more or less contradictory figures); this complexity determines the character’s “personality,” which is just as much a combination as the odor of a dish or the bouquet of a wine. The proper name... referring in fact to a body… draws the semic configuration into an evolving (biographical) tense.
-S/Z, XXVIII